Panic! At the Disco played a live rendition of their song "Don't Let the Light Go Out" tonight at the 2022 MTV VMAs. It was the fourth single from their latest album Viva Las Vengeance, and tonight served as the first time it was played live.
The band was among several other rockers who performed during the ceremony, including Red Hot Chili Peppers, who were also the recipients of the Global Icon Award, and Maneskin. Panic! At the Disco were nominated for Best Alternative Video for the song "Viva Las Vengeance," but Maneskin won the award with the video for "I WANNA BE YOUR SLAVE." Avril Lavigne, Machine Gun Kelly and Twenty One Pilots were among the other nominees for the category.
See the full list of rock and alternative winners here, and check out PATD's performance below. The story will be updated as more videos surface.
Panic! made their return earlier this summer when they dropped "Viva Las Vengeance" in June. The album of the same name came out last Friday (Aug. 19), and marked their first record since 2018's Pray for the Wicked, which was the band's second release to list frontman Brendon Urie as the only official member of the band. However, Urie worked with several musicians on Viva Las Vengeance, including Jake Sinclair, Mike Viola and Butch Walker.
Panic! At the Disco are heading out on a North American tour starting Sept. 8 in Austin, Texas. The run will wrap up Oct. 25 in San Francisco, and then they'll head over to Europe. See all of the dates on their website.
Watch Panic! At the Disco Perform at the 2022 MTV VMAs – Loudwire
Efficient enumeration-selection computational strategy for adaptive chemistry | Scientific Reports – Nature.com
Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Scientific Reports volume 12, Article number: 14664 (2022)
Metrics details
Design problems of finding efficient patterns, adaptation of complex molecules to external environments, affinity of molecules to specific targets, dynamic adaptive behavior of chemical systems, reconstruction of 3D structures from diffraction data are examples of difficult to solve optimal design or inverse search problems. Nature inspires evolution strategies to solve design problems that are based on selection of successful adaptations and heritable traits over generations. To exploit this strategy in the creation of new materials, a concept of adaptive chemistry was proposed to provide a route for synthesis of self-adapting molecules that can fit to their environment. We propose a computational method of an efficient exhaustive search exploiting massive parallelization on modern GPUs, which finds a solution for an inverse problem by solving repetitively a direct problem in the mean field approximation. One example is the search for a composition of a copolymer that allows the polymer to translocate through a lipid membrane at a minimal time. Another example is a search of a copolymer sequence that maximizes the polymer load in the micelle defined by the radial core-shell potentials. The length and the composition of the sequence are adjusted to fit into the restricted environment. Hydrogen bonding is another pathway of adaptation to the environment through reversible links. A linear polymer that interacts with water through hydrogen bonds adjusts the position of hydrogen bonds along the chain as a function of the concentration field around monomers. In the last example, branching of the molecules is adjusted to external fields, providing molecules with annealed topology, that can be flexibly changed by changing external conditions. The method can be generalized and applied to a broad spectrum of design problems in chemistry and physics, where adaptive behavior in multi-parameter space in response to environmental conditions lead to non-trivial patterns or molecule architectures and compositions. It can further be combined with machine learning or other optimization techniques to explore more efficiently the parameter space.
Finding a structure or a pattern among numerous possibilities that fits to a certain environment or performs a given function is an attractive target for predictive theoretical methods leading to profusion of scientific creativity. In general, problems of optimal design of materials that mimic, adapt or possess a controlled affinity to external conditions are examples of inverse problems1. Each inverse problem is associated with a direct problem, consisting in repetitively solving or simulating a well-defined model system for a concrete set of parameters and conditions2. These optimal design or inverse problems can be rather challenging2 and in real situations can be computationally expensive without additional conditions or regularization3, or require computationally expensive heuristic optimization methods, e.g. genetic algorithms4 and neural networks5,6.
The examples of such optimal design problems in chemistry and physics include the understanding of adaptiveness or binding affinity in the molecular recognition process within the lock-and-key paradigm7,8, which turns out to be very sensitive to molecular parameters9; the search for polymer sequence design in protein folding10, the interaction potentials for optimal self-assembly11, the search for dynamic exchange of components in response to environmental conditions12, the prediction of a sequence of cell-penetrating peptides13, inverse design problems in nanophotonics14,15.
A concept of Dynamic Combinatorial Chemistry (DCC) was introduced16 to address reversible connections and conformations of basic elements and chemical components that can rearrange and dynamically adapt to changes in the environment, which contrasts with static libraries of prefabricated molecules and compounds. Reversible and adaptive connections between elements can be grouped in virtual combinatorial libraries, while a search through libraries of connections in a changeable environment is another example of inverse problem that needs an efficient solution. This idea of experimentally creating adaptive molecules was proposed even before computational design algorithms were widely used to discover new materials.
Among computer simulation methods, one of them, the Reverse Monte Carlo (RMC) method17,18, produces three-dimensional structure of a molecular system that fits a given experimental data that originate from unknown molecular structure, such as, for example, diffraction and scattering data in the problems of inverse scattering or reconstruction from neutron diffraction patterns19. This is achieved by direct Monte Carlo simulation of these data from a structural model, in which parameters are adjusted in order to fit the experimental data. Since it is not practical in many situations to explore all possible combinations of parameters of the molecular structure unless the model is very simple, the resulting structure is not unique or exact and there is no guarantee that it produces the best fit of the data. Ideally, a method that exhaustively explores all parameter space tests all possible combinations would provide a guaranteed best solution.
Finding a solution to an inverse problem by exhaustively solving the direct problem in a complete parameter space resembles a brute-force password screening in computer science. However, in contrast to password generation, direct computer simulations can be very computationally expensive, thus making this approach too slow and almost impractical. Meanwhile, explosive growth20 of computer power provided by parallel computation on modern Graphical Processing Units (GPUs) can be game-changing in the field of inverse search problems, including direct sampling of the parameter space, thus providing a basis for adaptive chemistry basis on the level of computer simulations.
Nevertheless, many existing GPU implementations of Molecular Dynamics and Monte Carlo simulations are, in fact, adaptations of sequential codes to parallel environments, that cannot fully benefit from the potential speed-ups on modern parallel GPU architectures counting thousands of cores. Limitations in scalability come from the Amdahl law21 stating that even a small fraction of sequential code or unavoidable synchronization between cores would dramatically impact the scale-up. Hence, to get a full advantage of a parallel hardware architectures, conceptually new scientific methods that are fully parallel in nature have to be invented or re-discovered.
The architecture of a new generation of processors that are still to come dictates a completely different design of methods for solving the optimization problems. As an example, parallel architecture and explosive increase of computational power was the reason why neural networks, invented in the 80s were computationally too expensive to solve any real problem at that time and became so popular and efficient in the 00s. Despite being computationally more expensive than other direct methods, they scale well with the number of units and weights in the network permitting a large number of layers unless other stochastic methods, mentioned by the reviewer, that scale combinatorially22. This could be a similar trend with exact enumeration for inverse search problems in the near future when the number of parallel units would increase even more significantly, although it may be difficult to imagine it now, taking into consideration all limitations present in the modern parallel architectures such as limited shared memory and bandwidth of communication with the global memory. New methods should not only scale well with the number of units, but they should be perfectly scalable.
In the following, we report a theoretical method of solving optimal design problems using highly parallel technique taking advantage of massive parallelization on GPUs. After introducing the method, we demonstrate its performance on two examples: the search for a copolymer sequence that translocates through a lipid membrane at a minimal time and a copolymer sequence that maximizes the load by a copolymer micelle.
Top: Building blocks and topology of molecules that are created with the enumeration-selection strategy. Conformations are built from monomers connected by bonds and interacting with the fields: concentration fields or external fields. Various types of topology can be addressed. Bottom: three practical examples studied in detail: linear polymer with the composition adjusted to a lipid membrane translocation; linear polymer with hydrogen bonds adapting to the concentration gradient; branched polymer with a field of distributed branching points.
Inverse design workflow.
Exhaustive search through parameter space implies running massive simulations corresponding to a direct problem in efficient way. This is possible if these simulations are completely independent and do not require communications between the cores and thus would be a subject only to hardware limitations on the number of available cores. For example, such independence of tasks can be realized in the quasi-instantaneous field approximation, where individual molecules interact between each other through fixed or rarely updated mean fields23. It was shown that maintaining fields at most recent values and updating them at a certain frequency introduces correlations between the molecules. In such a case, the composition of a molecule, force fields and conformations of a single molecule can be strictly decoupled between each other, thus allowing to generate a molecule composition from building blocks, composition and force fields suitable for massive parallel implementation, Fig. 1. The workflow of the method is depicted in Fig. 2. Once the target parameters are identified: molecule properties and constraints, macroscopic material properties and the restrictions from external fields and conditions; the molecule is digitalized into building blocks (monomers), which are characterized by its size, connections and interactions. This object with annealed structure and parameters is placed in the fixed fields that are dictated by the design target. The fields can be physical fields or chemical gradients/concentrations. Then this adaptive molecule with unknown shape and structure needs to adjust to external conditions and the fields in the inverse enumeration of all conformations search, that can be combined with heuristic methods.
The concept of adaptability implies the external fields as a driving force for the change of the composition or the topology of the adaptive molecules. It can be a concentration fields of monomers, external physical fields, chemical gradients. The building blocks of this method are the monomers that have distinct colors representing interactions with the fields.
This concept is demonstrated in three distinct examples: (i) a linear copolymer of fixed length whose composition is adapted for translocation through a lipid membrane with a minimal time and the composition of a block copolymer in a micelle is optimized for the maximal load; (ii) adaptive linear polymer interacting with the surrounding through reversible links such as hydrogen bonding. The probability of forming a hydrogen bond depends on the concentration field surrounding the monomer; (iii) a branched molecule with annealed topology, where the probability of branching is influenced by the surrounding. Concentration fields of certain components can control the gelification or branching of the molecules, while the concentration fields can be tuned by external environments.
The method is similar in spirit to Single Chain Mean Field (SCMF) theory24 for polymers, where correlations between conformations of a single polymer chain are decoupled and each conformation of the polymer chain is interacting with mean fields independently. Soft matter objects such as micelles and lipid bilayers are, in fact, close to liquids, where molecules move independently and uncorrelated between each over. Thus, in most cases, such an approximation is valid and provides a good results for mechanical properties and self-assembly25. Since building blocks in this method are entire molecules with full sampling of conformations, the correlations between monomers inside the molecule are exactly calculated. Density Functional theory (DFT)26 has a similar spirit for atoms and small molecules. There are however particular cases, when this approximation is not valid: ion pairs, when two molecules are correlated and move together, movements in gel phase, crystallization, strong electrostatic interactions beyond mean field.
As an example of application of the method, let us consider a linear copolymer chain of a given length N and consisting of monomers of types a. The monomers of different types differ in interactions between each other and with external fields, which represent the environment. The aim of the method is to find a sequence of monomers that satisfy a given design criteria based on the fields, e.g. the affinity with the external fields. The sequence consists of monomers from the library, where a color corresponds to a given interactions with concentration and external fields, further denoted by (Phi). These monomers are connected between each other by freely joint bonds of fixed length, which can be represented by a cube on a cubic lattice. With such construction, each sequence of monomers can be in many conformational states, described by relative position of monomers in space.
One dimensional fields on a 3D cubic lattice corresponding to (A) planar geometry of a lipid membrane, (B) spherical geometry of a core-shell micelle. (C) Lattice representation of a copolymer sequence: black (P_T) and white ({P_H}) cubes represent hydrophobic and hydrophilic blocks, respectively. (D) Map of interactions between monomers (P_H), (P_T) and fields T, H and S.
We assume that each conformation (Gamma) corresponding to a given sequence is independent when placed in fixed fields (Phi _i(r)), where r is a point in space, Fig. 3. With this, the Hamiltonian of a conformation (Gamma) is described by a
where the integration is over the whole space, (mathscr {H}^{intra}(Gamma )) is the intra-molecular interaction energy resulting from interactions between monomers inside the same molecule; (U_i^{a}(Gamma ,r)) is the interaction potential between monomers of type a with the field i of conformation (Gamma) at the position r. With this, the total partition function of the system, Z, can be written as:
where (w_Gamma) are the weights of the conformational distribution, if it was generated with a bias27,28. Partition function Z provides all thermodynamic properties of the system, while the free energy is given by (mathscr {F}=-k_BTln Z), where (k_B) is the Boltzmann constant and T is the temperature.
Predicted sequences sorted according to their time of translocation through a lipid bilayer for a copolymer consisting of 12 monomers. Top: Mean escape time (tau) as a function of polymer chain sequence number sorted from fastest to slowest. Bottom: the three fastest and three slowest sequences. The color (from blue to red) corresponds to the average hydrophobicity, defined by the fraction of hydrophobic blocks (N_{T}/(N_{T}+N_{H})) and ranging from 1/12 to 11/12. For comparison, translocation times for homopolymers, 0/12 ((9.90 times 10^{10})), 12/12 ((4.96times 10^{12})) show two orders of magnitude difference.
Thus, by introducing independent conformations (Gamma) placed in fixed fields we provide a basis for a very efficient parallelization scheme, but for the price of neglecting correlations and limiting application to the mean field description. Each conformation (Gamma) of a given sequence can be generated independently on a single GPU core with strictly no communication with other cores. This can be realized with an efficient algorithm, such as reported GPU realization of the Rosenbluth algorithm29. The weights of the distribution (w_Gamma) and the Hamiltonian (H(Gamma)) can be calculated on-the-fly without a need to keep them in memory. The resulting (H(Gamma)) can be used for classification and ordering of the sequences according to a given design criteria. Thus, such an approach allows for a fast screening of the sequences using full power of parallel architecture of modern GPUs with no limitations of the Amdahl law21.
For numerical purposes, it is convenient to discretize the space and consider sums instead of space integrals in Eq. (1). We assume that the simulation box is divided into three fields as it shown in Fig. 3, hydrophobic field, (Phi _T), hydrophilic field, (Phi _H) and the solvent, (Phi _S). Thus, on a cubic lattice, where each monomer occupies one lattice site, the Hamiltonian, Eq. (1), is written as
where (U^{P_T}_H(Gamma ,x,y,z)) is the number of contacts of the conformation (Gamma) of the (P_T) polymer at the lattice site (x, y, z), interacting with the field (Phi _{H}). Similar definition is applied to (U^{P_H}_T(Gamma ,x,y,z)) and (U^{P_T}_S(Gamma ,x,y,z)), interacting with (Phi _{T}) and (Phi _{S}), correspondingly27,28. The lattice number (the maximum number of neighbors of a lattice site) is (z=26) for a cubic lattice. Thus, if one of these sites is occupied by a monomer T or H, or a solvent S, it is considered as a contact with a corresponding energy (varepsilon).
The incompressibility condition is imposed by the requirement that each lattice site can only be occupied either by a monomer or by a solvent. The lattice site, occupied by a polymer does not interact with other monomers. We assume that all units, (P_T), (P_H), T, H and S, have the same size of one lattice unit and the same interaction range (nearest neighbors only) and differ only in corresponding interaction energies: (epsilon_{P_T-H}=0.1) (k_BT), (epsilon _{P_H-T}=0.1) (k_BT), (epsilon _{P_T-S}=0.1) (k_BT). All other interactions are set to zero (Fig. 3C and D). They are chosen in such a way that if the polymer chain is composed entirely of (P_T), it would be fully hydrophobic, and if it is composed entirely of (P_H), it would be fully hydrophilic.
Statistical segments of chain-like molecules are represented as connected unit cubes on a simple cubic lattice. The distance between bonded monomers is constrained to a bond vector set of 26 bond vectors by (1,0,0),(1,1,0),(1,1,1), for each monomer, interacting maximum with the 26 nearest neighbor sites. The excluded volume is implemented via the exclusion of overlap between monomer on the same lattice site, while the distances are expressed in terms of the number of lattice sites. The simulation details and the algorithm can be found in Supplementary Materials.
The calculation efficiency depends mainly on the polymer length and sampling of conformations. The following estimates were obtained for a machine equipped with NVIDIA Tesla V100 GPU and Intel Xeon Gold 6248 CPU. The calculation of the translocation time through a lipid membrane in the planar geometry, typical simulation time for a polymer chain of length (N=12) is (0.0384 pm 0.0004) s per million of conformations and for a chain of length (N=16) is (0.0784 pm 0.0004) s per million of conformations. This typical time allows for the generation of a significant statistical sampling for each pattern/sequence. For the spherical geometry of a core-shell micelle, the typical simulation time for the polymer of length (N=16) is similar to planar geometry: (0.0834 pm 0.0003) s per million of conformations.
In this section we describe linear polymers which composition (sequence of monomers) is adapted to a given design criteria. Two examples are considered: the problem of optimization of passive translocation time through a lipid membrane represented by one dimensional energy barriers, and the problem of confined and constrained polymers on the example of the localization of a linear polymer in a concentric field of a spherical micelle.
The problem of finding the composition of translocating polymer that minimizes the time of diffusive translocation through a lipid bilayer was studied in detail for homopolymers30,31,32 and amphiphilic copolymers33. Thus, we use these results as a reference to benchmark the predictions and the results of the enumeration-selection strategy.
A lipid membrane is represented by a one dimensional hydrophobic field of lipid tails (Phi _{T}) with a thickness of 6 lattice units, surrounded by a hydrophilic field of lipid heads (Phi _{H}) of two lattice units as shown in Fig. 3A. The rest of the box is filled by the solvent (Phi _{S}). We take for the measure of the design criteria the translocation time of a polymer with given sequence through the lipid membrane. The mean first escape time, (tau),
is defined between (z_-) and (z_+) designating the position of the top and the bottom of the simulation box, correspondingly. Here, (mathscr {F}(z)) is the free energy of a layer z of conformations (Gamma), which have the center of mass, (bar{z}(Gamma )), lying at a distance z from the bilayer’s mid-plane:
Here, (w_Gamma) is the bias of the distribution (in case of Rosenbluth generation it is a Rosenbluth weight34), and (mathscr {H}(Gamma )) is the Hamiltonian of a configuration, Eq. (3). In the following, the lattice constant (1) is used. It is important to note, that the polymer length-dependent diffusion constant is not considered in the expression (4) and thus, the following results represent timescales with respect to the self-diffusion time.
For simplicity of understanding the method, we assume that the monomers in a linear polymer chain can be of two types: hydrophobic ((P_T), black) and hydrophilic ((P_H), white). More types of monomers, branching of polymers and even gradual change between hydrophobic and hydrophilic monomers are possible extensions of the model. Each polymer sequence or pattern is thus a sequence of monomers of two kinds. The polymer properties are expected to be invariant under reversion of the sequence between both ends.
(A) Predicted fastest patterns (Min) and slowest patterns (Max), according to the mean escape time (tau) for a fixed ratio of hydrophobic (black) and hydrophilic (white) monomers, (N_{T}/(N_{T}+N_{H})=1/2). Shown up to length (N=16), while similar alternating patterns are observed for longer polymers (up to (N=24), not shown for clarity). (B) The corresponding mean escape time (tau) of the slowest and fastest patterns as a function of the polymer length N for a fixed ratio, (N_{T}/(N_{T}+N_{H})=1/2).
At a given ratio between hydrophobic and hydrophilic monomers, (N_{T}/(N_{T}+N_{H})), the total number of possible sequences is (N!/({N_{T}!N_{H}!)}), where (N=N_{T}+N_{H}). It grows exponentially with the length of the polymer N. For each polymer sequence the exhaustive search method consists of the sampling of (~10^7) independent self avoiding random walks that are uniformly distributed in the box and thus, samples sufficiently precise the free energy (mathscr {F}) (Eq. (4)) of conformations at each position z. Technically, the conformations are not stored in memory, but are only used to sample the free energy in parallel on GPU on-the-fly such that billions of conformations can be generated in a reasonable time.
During the generation process, the mean escape time (tau) is calculated taking into account all generated conformations. The process is performed for each sequence and each polymer length up to (N=24). As a result, sequences can be ranked with respect to the time (tau).
The results for polymer length (N=12) are presented in Fig. 4 and in Table S1. It shows that the translocation time is very sensitive to the sequence of the polymer. The most efficient translocation is found for the balanced hydrophobicity, (N_{T}/N=6/12). This is consistent with the previous systematic findings using Monte Carlo simulations30,31. The difference of the mean escape time between different sequences is several orders of magnitude. At the balanced point, sequences with shortest hydrophilic/hydrophobic blocks show smallest translocation times only due to weak localization at the bilayer-solvent interfaces as compared to sequences with larger blocks. In accordance with a recent work35, we also see in Fig. 4 for slightly hydrophilic sequences, (N_{T}/N<6/12), that formation of larger hydrophobic blocks near the center of the sequence leads to the smallest translocation times in this case.
Mean escape times show equivalent behavior as a function of the sequence for the various chain lengths investigated as shown in Fig. 5. There, the fraction of hydrophobic units is fixed to 1/2 for various polymer lengths. The results demonstrate that the sequences with alternating hydrophobic and hydrophilic monomers provide the fastest translocation time at balanced hydrophobicity. Both maximum and minimum data for (tau (N)) corresponding to alternating and diblock sequences can be approximated by an exponential increase with N (Fig. 5b). While there is only marginal increase of translocation time for alternating sequences which only weakly interact with the given diffusion barrier, for the slowest sequences (diblock copolymers) the slope is significantly larger, see Fig. 5b and Table S2. Assuming that the rate limiting process is the detachment of the hydrophobic block into solvent, the contact energy contribution is written as (z_{eff}epsilon _{P_T-S} (N/2)) with (z_{eff}) being an effective coordination number counting the number of lattice contacts of (P_T) monomers with solvent. The ratio (sim 10^{7}) between the maximum translocation time for (N=24) (Fig. 5a), and the corresponding free diffusion result (tau sim |2z_pm |^2sim 10^3) reflects an excess free energy barrier of (F_{barrier}=18k_BT) for desorption of the diblock. An effective coordination number is then estimated as (z_{eff}approx 15), which is slightly smaller as compared to the result (z_{eff}approx 19.6) given in Ref.35. There, also contributions from conformational entropy change upon desorption as well as much shorter chains have been considered only. A more precise consideration will also take into account conformational entropy of confinement of the (P_T)-block into a slit-geometry given by the membrane thickness.
In this example, we address the problem of finding an optimal polymer sequence that can fit the best to a core-shell structure. It can be a problem of a maximum load of a block-copolymer micelle with a polymer or a packing problem of a polymer in a field of concentric layers. More generally, this can be an inverse problem of finding the molecular structure of the compound that creates a given field. In the following, we describe an example of inverse localization problem, i.e. a problem of finding a linear polymer sequence that is at a given position in space.
The localization problem of the given (purple) monomers in two concentric layers: hydrophobic core (red) and hydrophilic corona (blue). On the left: a schematic illustration of the criteria, on the right: three most probable and three less probable sequences obeying the criteria. Black is hydrophobic and white is hydrophilic monomers. (A) Central monomer (i=8) should be localized in the center of the micelle, (0<r<2), boundaries sketched as D– and D+; (B) polymer ends (i=1) and (i=16) should be localized at the border between the core and the corona, (10<r<13); (C) Polymer ends (i=1) and (i=16) should be localized in the corona, (13<r<16); (D) All monomers (i=1, dots ,16) should be localized in the corona (13<r<16).
We consider spherical symmetry in such a way that a given field represents concentric core-shell layers and we impose that a given monomer in the chain (hydrophilic or hydrophobic) has a maximum concentration in the chosen layer. To formulate this criteria for the localization problem, we define average concentrations of the i-monomer (bar{C}_i(x,y,z)) at a given position in space (C_i(x,y,z)) as
where (mathscr {H}(Gamma )) is the effective Hamiltonian, defined in Eq. (3), Z is the normalization constant, Eq. (2), and (C_i(x,y,z)) is the concentration of the i-monomer of a conformation (Gamma) at a given point on space (x, y, z).
With a radial symmetry, (x, y, z) is reduced to concentric layer of radius r and the layer thickness (sigma). Additionally, we can fix the number of monomers (Mle N) to be located in the concentric layer. The design criteria for the localization of i-monomer in the concentric layer (D=(r,r+delta )) can thus be formulated as the maximum of the fitting function (theta _{MD})
To illustrate this concept, we consider a concrete example of the micelle given by two concentric fields, hydrophobic in the core and hydrophilic in the corona. We fix the concentric field since we impose a polymer composition to adjust to that field. The field is defined as a hydrophobic layer with the radius (L_{T} = 13) lattice units in the center and the hydrophilic shell of thickness (L_{H} = 3) lattice units. The composition of amphiphilic copolymer of length (N=16) with equal ratio between hydrophobic and hydrophilic monomers (hydrophobicity 50%), have to be adjusted to the given field in different variations, Fig. 6.
For such a polymer, the total number of sequence combinations is 12870. The predicted sequences that obey the localization criteria are sorted according to (theta _{MD}). Imposing different monomer localization criteria, the resulting sequence pattern can be very different. A) A central monomer (i=8) of the sequence M (purple) should be localized in the core of the core-shell structure ((0<r<2)). The best sequences correspond to strictly alternating copolymers, where the polymer passes through the core, thus allowing the central monomer to be located in the middle. Tri-block copolymers with hydrophobic ends are the worse rated according to this criteria. B) Monomers on both ends ((i=1) and (i=16)) should be localized at the the border between hydrophobic and hydrophilic regions ((10<r<13)) of a micelle. In this case the most adapted sequence is the the triblock polymer with two hyrdophilic ends and hydrophobic block in the middle. C) The ends should be located in the hydrophilic region ((13<r<16)) of the micelle. A tri-block copolymer with hydrophilic ends is the most adapted sequence. D) The whole polymer should be located in the hydrophilic region ((13<r<16)). Tri-block copolymer with a amphiphilic block in the middle and two hydrophobic ends are the most adapted sequences, while alternating copolymer is the less adapted.
In this section we introduce more complex molecular structures such as reversible links between molecules and hierarchical branching of molecules.
The increased complexity of the topology of the molecules allows for more degrees of freedom for the molecular design and provides a flexibility for adapting the molecule structure and composition to the changes of the environment in the accordance with the concept of dynamic combinatorial libraries for supra-molecular chemistry36. The range of complex structures that can be obtained with this strategy have received the name “dynamers”37,38 and some of them are depicted in Fig. 1. For example, reversible and dynamic links can be formed by hydrogen bonds or by recognition-directed reversible polyassociation through complementary interaction groups or reversible polycondensation and reversible functional groups. Classical example of a simple reversible system is a linear living polymer39. These various types of the reversibility can be implemented in a mean field through effective interactions between monomers and the solvent, while the average composition, length distributions, etc, are determined by the thermodynamic equilibrium. Shifting the equilibrium by changing external conditions leads to selectivity and responsiveness of a reversible self-assembled system.
One of the form of reversible links is hydrogen bonds either between components or between monomers and water molecules. One of well-known examples is a polyethylene oxide (PEO) in water, which is water-soluble due to the formation of dynamic hydrogen bonds with water40,41,42. Hydrogen bonding can be described within a two-state model, where a monomer can be in two states: a monomer forming a hydrogen bond with water and a monomer without a hydrogen bond43. These monomers have different interaction parameters with water, while their position and number are determined from chemical equilibrium. The origin of two states is the gauche-trans equilibrium of monomer conformations that have different probabilities to form a hydrogen bond with water. Higher polymer concentration makes it more difficult to form a hydrogen bond, thus the probability to be in one or another state is concentration dependent.
Dynamic formation of hydrogen bonds between monomers and water of a linear PEO chain. (A) A model of a two-state polymer, where black monomers correspond to the formation of a hydrogen bond with wate and white monomers correspond to no bond. The probability to have a bond is coupled with the polymer concentration. (B) Most probable configuration in the field of a polymer concentration. (C) Polymer concentration and the fraction of hydrogen bonds (orange) versus the distance from the center r of a two-state polymer in the polymer concentration field (blue). Dashed lines correspond to a sparse globule; (D) Distribution of block sizes blocks in the concentration field and their relative frequency, (f=f_{HB}/(f_{HB}+f_{no-HB})). The corresponding H-bond distribution is proportional to solvent volume fraction.
As an example, consider a linear polymer consisting of (N=16) monomers, where white monomers correspond to hydrogen bonding (hydrophilic) and black blocks correspond to free monomers (hydrophobic). The fraction of blocks is fixed to 50% and controlled by external conditions, but the position is not fixed and is coupled with the polymer concentration: (phi _p=1.0 – 1.0/(1.0+exp (-r+9.0))) corresponds to a dense globule, solid line in Fig. 7C and (phi _p=0.5 – 0.5/(1.0+exp (-r+9.0))) for a sparse globule, dashed line. In turn, the conformation with exact sequence of white and black blocks, depends on the distribution of the hydrophilic and hydrophobic blocks along the chain, Fig. 7. Running the search for a composition of a polymer in its own concentration field in a radial symmetry leads to the following results: the fraction of hydrogen bonds increases as the polymer density decreases with the distance from the center. If the density gradient of a polymer concentration is shifted, for example, due to changing external conditions such as temperature or pH, the fraction of hydrogen bonds would adjust to the field and the hydrogen bonds would re-distribute to adapt to the field.
Often molecules are not linear and can have a complex topology such as branches or side groups. Branching provides an additional degree of freedom for the design of adaptive polymers. Considering the probability of branching (theta) for each generated monomer during Rosenbluth generation, different branching topologies can be obtained, Fig. 8. With this, the total number of branching points and the functionality of branches (number of branches at the branching point) can be controlled through external fields.
Hierarchical branching in Rosenbluth sampling is introduced as follows: the first monomer is placed at a random position in the simulation box. Next monomers are added one after another following a classic Rosenbluth chain growth algorithm and checking their self-avoidance. Starting from the third monomer, the polymer can branch with a given probability of branching (theta) which is checked according usual Monte Carlo procedure. The branching functionality f is introduced as the number of branches per branching point. Each branch is then followed in a recursive way and the random chain growth process continue until it reaches the given total polymer length. The resulting polymer provides the distribution of branches that follows the total density distribution, Fig. 8.
Such an approach allows to create molecules with adaptive topology that is controlled by external conditions. Examples could be polyimines37,38 derived from various diamines and dialdehydes that are sensitive to changes in absorption and fluorescence spectra as well as in their solubility in organic and aqueous media.
Adaptive branching of the molecules in the concentration field. (A) A model of a branched polymer, where black cubes designate hydrophobic monomers and white cubes designate hydrophilic monomers. The branching point is designated with two bonds to branches. (B) Most probable configuration of the branching polymer with a given probability of branching coupled with the field of polymer concentration. (C) Polymer concentration (blue) and the distribution of branching points of an adaptive polymer as a distance from the center r, with a given probability of branching (theta =0.2). Dashed lines correspond to a sparse globule. (D) Example of a 3D structure of a branched polymer conformation used in the study.
Exact enumeration-selection computational strategy is a powerful tool for relatively small sets of conformation space. If the number of the building blocks increases, the number of conformations increases rapidly with the length N and generation of all conformations becomes impractical. One of the strategy is to use representative sampling, i.e. large enough sampling to ensure that the averages statistically do not depend on sampling, instead of a full sampling. This is also the solution for off-lattice models, where the method is readily generalized the same way as a standard SCMF theory with a representative sampling27. Further, the enumeration method can be combined and complemented with heuristic methods for efficient search in the parameter space such as neural networks (NNs), genetic algorithms or any other method for exploration in large parameter space. This combination still relies on large conformational sampling to properly sample the statistical conformations of a single chain.
To illustrate such coupling and complementarity of the methods, a special neural network was designed to learn the relation between the structure (block copolymer sequence) and the translocation time through the membrane (target fitting function) with the same model parameters as we used for exact enumeration for the problem of finding a polymer sequence for optimal translocation through the membrane (first example). The method and the structure of the network is described in Ref.35.
The comparison allows to reveal the fundamental differences in the approaches and highlight the conceptual strengths and weaknesses of both approaches that can be combined. The enumeration-selection strategy implies direct exploration of the whole conformational and parameter space. This gives exact answer what is the optimal composition without the risk to be trapped in subset of parameters corresponding to a local minimum. In turn, NN provides the prediction for optimal solution based on training set and entirely depends on how large and complete this set is. Machine learning does not guarantee that the predicted solution is the true optimal solution corresponding to the global minimum, thus often called a black box without possibility to verify the predicted results. However, NNs are generally efficient in exploration of large sets of parameters, while the time of computation for achieving the optimal solution with enumeration strategy increases exponentially with the size of the system and number of parameters to explore, limiting its applicability. Thus, a combination of machine learning with exact enumeration-selection strategy with a balanced sampling generated by exact enumeration for training of NN could be the best strategy for solving design problems.
Evaluation of the performance of the hybrid prediction method, based on convolutional neural network (CNN) with exact enumeration used as a training set. (delta) is the mean square error (target precision) and (nu _{clock}) is the number of processor clock cycles multiplied by the number of parallel threads employed in units of 1012 cycles.
The comparison between the performances of enumeration-selection and neural network methods is demonstrated on the example of a sequence of an amphiphilic polymer of length (N=14) monomers. In particular, the performance of NN depends on the size of the training set. This training set (training window) is generated by exact enumeration for which the result is known exactly. The prediction of NN is evaluated on a predicted set, while the performance of NN is defined as a ratio between predicted and training sets. NN can correctly predict much larger sets for the price of precision. In this example, the targeted precision for translocation time (tau) is given as a mean squared error (MSE) (delta) of the fitting function (log tau), while the computational efficiency is measured as ((1/delta )times (1/nu _{clock})), where (nu _{clock}) is the total number of (parallel) clock cycles (in Tera-clocks) spent for the prediction, Fig. 9. The performance is evaluated as the fraction between predicted sequences to sequences used for training (sampling set obtained by exact enumeration).
Three conclusions can be made from the comparison with NN method: (i) A high precision of machine learning prediction can only be achieved by a large training set, which can be generated by exact enumeration, and as a counterexample, the yellow curve is larger than the red or pink or green(s) curves in Fig. 9. (ii) For a given precision, the NN reduces the computational effort most for the higher precision. (the slope of the orange curve is smaller than that of the blue). (iii) Computational efficiency of a hybrid method, which includes exact enumeration of the training set and NN for prediction for “unseen” conformations set, relies on the size of the training set. For any given precision, a higher fraction of NN contribution reduces the computational cost.
In brief, the exact enumeration can be used as a ground truth for creation of the training set for training the neural network, while machine learning can extend the method for exploration of large systems and large parameter space (heuristic exploration), thus providing a powerful combination of inter-related complementary methods.
We presented exhaustive search method for design of composition of polymer sequences in the mean field approximation. This method allows to solve inverse (or design) problem with brute-force solution of direct problem. The key of performance is massive parallelization on GPUs which almost eliminates communication between cores which makes this method significantly more efficient than existing computationally expensive heuristic optimization methods such as deep learning, neural networks or genetic algorithms. The performance of the method was demonstrated on two examples: polymer sequence that translocate through lipid bilayer with a minimal time and the the sequence with best affinity to spherical core-shell structure, representing a micelle. This method is readily generalized to wide spectra of inverse (or design) and combinatorial search problems in chemistry and physics.
All data generated or analyzed during this study are included in this published article and its supplementary information files.
Inverse Problems in Engineering Mechanics III (Elsevier, 2002). https://linkinghub.elsevier.com/retrieve/pii/B9780080439518X50000.
Stanley, M. B., Osher, S. J. & Yablonovitch, E. Inverse problem techniques for the design of photonic crystals. IEICE Trans. Electron. 87, 258–265 (2004).
Google Scholar
Benning, M. & Burger, M. Modern regularization methods for inverse problems. Acta Numerica 27, 1–111 (2018). https://www.cambridge.org/core/journals/acta-numerica/article/modern-regularization-methods-for-inverse-problems/1C84F0E91BF20EC36D8E846EF8CCB830.
Delsanto, S., Griffa, M. & Morra, L. Inverse Problems and Genetic Algorithms. In Delsanto, P. P. (ed.) Universality of Nonclassical Nonlinearity, 349–366 (Springer New York, New York, NY, 2006). http://link.springer.com/10.1007/978-0-387-35851-2_22.
Adler, J. & Öktem, O. Solving ill-posed inverse problems using iterative deep neural networks. Inverse Problems 33, 124007 (2017). http://stacks.iop.org/0266-5611/33/i=12/a=124007.
Kim, K. et al. Deep-learning-based inverse design model for intelligent discovery of organic molecules. npj Computational Materials 4, 67 (2018). https://www-nature-com.sabidi.urv.cat/articles/s41524-018-0128-1.
Behr, J.-P. (ed.) The lock-and-key principle: The state of the art–100 years on. No. v. 1 in Perspectives in supramolecular chemistry (Wiley, Chichester [England] ; New York, 1994).
Conrad, M. Molecular computing: The lock-key paradigm. Computer 25, 11–20 (1992).
Article Google Scholar
Thalmann, F. A schematic model for molecular affinity and binding with Ising variables. Eur. Phys. J. E 31, 441–454. https://doi.org/10.1140/epje/i2010-10600-9 (2010).
CAS Article PubMed Google Scholar
Yue, K. & Dill, K. A. Inverse protein folding problem: Designing polymer sequences. Proc. Natl. Acad. Sci. USA89, 4163–4167 (1992). https://www.ncbi.nlm.nih.gov/pmc/articles/PMC525653/.
Torquato, S. Inverse optimization techniques for targeted self-assembly. Soft Matter5, 1157 (2009). http://xlink.rsc.org/?DOI=b814211b.
Osypenko, A., Dhers, S. & Lehn, J.-M. Pattern generation and information transfer through a liquid/liquid interface in 3D Constitutional dynamic networks of imine ligands in response to metal cation effectors. J. Am. Chem. Soc. 141, 12724–12737. https://doi.org/10.1021/jacs.9b05438 (2019).
CAS Article PubMed Google Scholar
Holton, T. A., Pollastri, G., Shields, D. C. & Mooney, C. CPPpred: Prediction of cell penetrating peptides. Bioinformatics29, 3094–3096 (2013). https://academic.oup.com/bioinformatics/article/29/23/3094/246449.
Molesky, S. et al. Inverse design in nanophotonics. Nature Photonics12, 659 (2018). https://www-nature-com.sabidi.urv.cat/articles/s41566-018-0246-9.
Pilozzi, L., Farrelly, F. A., Marcucci, G. & Conti, C. Machine learning inverse problem for topological photonics. Commun. Phys. 1, 57 (2018). https://www-nature-com.sabidi.urv.cat/articles/s42005-018-0058-8.
Lehn, J.-M. Dynamic Combinatorial Chemistry and Virtual Combinatorial Libraries. Chem. A Eur. J.5, 2455–2463 (1999). https://onlinelibrary.wiley.com/doi/abs/10.1002/%28SICI%291521-3765%2819990903%295%3A9%3C2455%3A%3AAID-CHEM2455%3E3.0.CO%3B2-H.
McGreevy, R. L. & Pusztai, L. Reverse Monte Carlo Simulation: A new technique for the determination of disordered structures. Mol. Simul. 1, 359–367. https://doi.org/10.1080/08927028808080958 (1988).
Article Google Scholar
McGreevy, R. L. Reverse Monte Carlo modelling. J. Phys.: Condens. Matter 13, R877 (2001). http://stacks.iop.org/0953-8984/13/i=46/a=201.
Playford, H. Y., Owen, L. R., Levin, I. & Tucker, M. G. New insights into complex materials using reverse Monte Carlo modeling. Annual Review of Materials Research 44, 429–449 (2014). https://doi.org/10.1146/annurev-matsci-071312-121712.
Nobile, M. S., Cazzaniga, P., Tangherloni, A. & Besozzi, D. Graphics processing units in bioinformatics, computational biology and systems biology. Brief Bioinform 18, 870–885 (2017). https://academic.oup.com/bib/article/18/5/870/2562773.
Hill, M. D. & Marty, M. R. Amdahl’s Law in the Multicore Era. Computer 41, 33–38 (2008).
Article Google Scholar
Poggio, T., Banburski, A. & Liao, Q. Theoretical issues in deep networks. PNAS 117, 30039–30045 (2020). https://www.pnas.org/content/117/48/30039. Publisher: National Academy of Sciences Section: Colloquium on the Science of Deep Learning.
Daoulas, K. C. & Muller, M. Single Chain in Mean Field Simulations: Quasi-Instantaneous Field Approximation and Quantitative Comparison with Monte Carlo Simulations. J. Chem. Phys. 125, 184904 (2006).
ADS Article Google Scholar
Szleifer, I., Ben-Shaul, A. & Gelbart, W. M. Chain organization and thermodynamics in micelles and bilayers. II. Model calculations. J. Chem. Phys. 83, 3612 (1985). http://link.aip.org/link/JCPSA6/v83/i7/p3612/s1&Agg=doi.
Muller, M., Katsov, K. & Schick, M. Biological and synthetic membranes: What can be learned from a coarse-grained description? Phys. Rep. 434, 113–176 (2006). http://linkinghub.elsevier.com/retrieve/pii/S0370157306002924.
Jones, R. Density functional theory: Its origins, rise to prominence, and future. Rev. Mod. Phys.87, 897–923 (2015). https://link.aps.org/doi/10.1103/RevModPhys.87.897.
Pogodin, S. & Baulin, V. A. Equilibrium insertion of nanoscale objects into phospholipid bilayers. Curr. Nanosci. 7, 721–726 (2011).
ADS CAS Article Google Scholar
Pogodin, S. & Baulin, V. A. Coarse-grained models of phospholipid membranes within the single chain mean field theory. Soft Matter 6, 2216–2226 (2010).
ADS CAS Article Google Scholar
Guo, Y. & Baulin, V. GPU implementation of the Rosenbluth generation method for static Monte Carlo simulations. Comput. Phys. Commun.216, 95–101 (2017). http://www.sciencedirect.com/science/article/pii/S0010465517300929.
Sommer, J.-U., Werner, M. & Baulin, V. A. Critical adsorption controls translocation of polymer chains through lipid bilayers and permeation of solvent. EPL (Europhysics Letters)98, 18003 (2012). http://stacks.iop.org/0295-5075/98/i=1/a=18003?key=crossref.f866cb2ca6bf60847686dcf7db2acbfe.
Werner, M., Sommer, J.-U. & Baulin, V. A. Homo-polymers with balanced hydrophobicity translocate through lipid bilayers and enhance local solvent permeability. Soft Matter8, 11708 (2012). http://xlink.rsc.org/?DOI=c2sm26008e.
Werner, M., Bathmann, J., Baulin, V. A. & Sommer, J.-U. Thermal Tunneling of Homopolymers through Amphiphilic Membranes. ACS Macro Lett. 247–251 (2017). http://pubs.acs.org/doi/abs/10.1021/acsmacrolett.6b00980.
Werner, M. & Sommer, J.-U. Translocation and induced permeability of random amphiphilic copolymers interacting with lipid bilayer membranes. Biomacromol 16, 125–135. https://doi.org/10.1021/bm501266x (2014).
CAS Article Google Scholar
Rosenbluth, M. N. & Rosenbluth, A. W. Monte Carlo calculation of the average extension of molecular chains. J. Chem. Phys. 23, 356–359 (1955).
ADS CAS Article Google Scholar
Werner, M., Guo, Y. & Baulin, V. A. Neural network learns physical rules for copolymer translocation through amphiphilic barriers. npj Comput. Mater.6, 72 (2020). https://www.nature.com/articles/s41524-020-0318-5. Number: 1 Publisher: Nature Publishing Group.
Fujii, S. & Lehn, J.-M. Structural and functional evolution of a library of constitutional dynamic polymers driven by Alkali Metal Ion recognition. Angewandte Chemie Int. Edition 48, 7635–7638. https://doi.org/10.1002/anie.200902512 (2009).
CAS Article Google Scholar
Lehn, J.-M. Dynamers: Dynamic molecular and supramolecular polymers. Progress Polym. Sci. 30, 814–831 (2005). http://www.sciencedirect.com/science/article/pii/S0079670005000730.
Lehn, J.-M. Dynamers: From supramolecular polymers to adaptive dynamic polymers. In Percec, V. (ed.) Hierarchical Macromolecular Structures: 60 Years after the Staudinger Nobel Prize I, vol. 261, 155–172 (Springer International Publishing, Cham, 2013). http://link.springer.com/10.1007/12_2013_267.
Cates, M. E. & Candau, S. J. Statics and dynamics of worm-like surfactant micelles. J. Phys.: Condens. Matter 2, 6869–6892 (1990). https://doi.org/10.1088%2F0953-8984%2F2%2F33%2F001 Publisher: IOP Publishing.
Karlstroem, G. A new model for upper and lower critical solution temperatures in poly(ethylene oxide) solutions. J. Phys. Chem. 89, 4962–4964. https://doi.org/10.1021/j100269a015. Publisher: American Chemical Society (1985).
Dormidontova, E. Role of competitive PEO-water and water-water hydrogen bonding in aqueous solution PEO behavior. Macromolecules 35, 987–1001 (2002).
ADS CAS Article Google Scholar
Bekiranov, S., Bruinsma, R. & Pincus, P. Solution behavior of polyethylene oxide in water as a function of temperature and pressure. Phys. Rev. E 55, 577–585 (1997).
ADS CAS Article Google Scholar
Baulin, V. A. & Halperin, A. Concentration dependence of the flory chi parameter within two-state models. Macromolecules 35, 6432–6438 (2002). http://pubs.acs.org/doi/abs/10.1021/ma020296o.
Download references
The authors gratefully acknowledge Prof. Jean-Marie Lehn for valuable comments that improved the manuscript. VAB acknowledges financial assistance from the Ministerio de Ciencia, Innovación y Universidades of the Spanish Government through research project PID2020-114347RB-C33, financed by MCIN/ AEI 10.13039/501100011033. YG acknowledges funding from Natural Science Foundation of Jiangsu Province (BK20221437), Science and Technology Innovation 2030 (2021ZD0201301) and Academic Consortium 21 (SPF).
Kuang Yaming Honors School, Nanjing University, Nanjing, 210023, China
Yachong Guo
Leibniz-Institut für Polymerforschung Dresden e.V., Hohe Strasse 6, 01069, Dresden, Germany
Marco Werner
Departament Química Física i Inorgànica, Universitat Rovira i Virgili, Marcel.lí Domingo s/n, 43007, Tarragona, Spain
Vladimir A. Baulin
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
Y.G. performed the calculations of the pattern search; M.W. performed the benchmark with neural network; V.A.B. designed the research. All authors contributed with writing and reviewed the manuscript.
Correspondence to Vladimir A. Baulin.
The authors declare no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and Permissions
Guo, Y., Werner, M. & Baulin, V.A. Efficient enumeration-selection computational strategy for adaptive chemistry. Sci Rep 12, 14664 (2022). https://doi.org/10.1038/s41598-022-17938-x
Download citation
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-022-17938-x
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
Advertisement
Advanced search
© 2022 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.
Howling Music Names Marc Mannino as Executive Producer, West Coast – SHOOT Online
Marc Mannino has joined Howling Music as Executive Producer, West Coast. The veteran music industry executive, who most recently served as executive producer at San Francisco-based Swell Music + Sound, will spearhead new business development for West Coast advertising agencies, brands and other music clients.
Mannino brings broad experience in business development, strategic partnerships and music licensing. “Marc’s expertise will be a huge asset as we continue to build relationships and broaden our profile,” said Howling Music creative director Ryan Claus. “He knows about a lot of stuff we’ve never even heard of.”
For his part, Mannino says he’s a longtime fan of Howling Music, both as an original music house and burgeoning music library. “They’re surprising,” he says. “I thought these guys were mostly insane or heavily medicated, but it turns out they’re actually just really good at what they do.”
At Swell, Mannino oversaw original music production, sound design and sound mixing for brands, including Credit Karma, Facebook, Fitbit, Square, Golden State Warriors, Uber, Lyft, and others. His previous experience includes two years as director of music licensing and sales at Elias Arts, focused on library and original music production, music supervision and sonic branding for East Coast accounts. He began his career in product development at Atlantic Records.
“I am excited by the opportunity to build a solid West Coast presence for Howling Music,” Mannino says. “These guys will either completely shake up the music-for-advertising world or will wind up in jail. I’d give them even odds at either.”
Howling Music is based in Nashville.
For more information, visit www.howlingmusic.com
Howling Music, a music production company offering original music and music licensing for advertising, branded content and film and television, today announces the release of Till We’re Minted, an EP from Nashville-based Hip-Hop artist…
Howling Music scored the music for a new ad from the NFL, the Trevor Project and Los Angeles agency 72andSunny. Released in tandem with Pride Month, the spot proclaims the league’s support for the LGBTQ+ community through series of statements in…
Howling Music has hired agency producer Brittany Gutierrez to join the Nashville-based music house as Executive Producer. Gutierrez brings over a decade of producer experience, most recently with Ford’s longtime key agency GTB. Her focus will be…
Atlanta-based rap and hip hop artist KC Carter takes listeners on a deeply personal musical journey exploring sacrifice and success in The Come Up, a dynamic collection of 12 new songs, now available for download on all platforms via…
The Marine Corps marked its 244th anniversary in celebratory manner with the release of an ad campaign featuring its first-ever “birthday” running cadence. Conceived by Atlanta agency Wunderman Thompson and featuring a soundtrack…
The Trusted Source For News, Information, Industry Trends, New ScreenWork, and The People Behind the Work in Film, TV, Commercial, Entertainment Production & Post Since 1960.
Master the one-on-one meeting – Journal of Accountancy
This site uses cookies to store information on your computer. Some are essential to make our site work; others help us improve the user experience. By using the site, you consent to the placement of these cookies. Read our privacy policy to learn more.
AICPA:
When it comes to leading others, the one-on-one meeting is an opportunity for managers to get to know their direct reports as individuals while gleaning the information needed to guide them in the short and long term. But not every manager receives an instruction manual on how to conduct these meetings. A few simple tips can help managers make the most out of the one-on-one for themselves and their direct reports.
David Almonte, CPA, CGMA, might love being a boss to 15 "rock stars," but he feels slightly nervous before the one-on-one meetings he holds with each of them.
For Almonte, a financial reporting and analysis senior manager with Amica Insurance in Rhode Island, the jitters are a good thing. They mean "that I truly care about why I am there and what I am doing," he said. "It is a reminder that the conversation I am about to have or give is meaningful to me."
When done right, these meetings can be an extremely rewarding aspect of management. "At the end of the day, the best part of my job is helping other individuals grow and succeed in their careers," said Almonte, a winner of the 2022 AICPA Outstanding Young CPA Award.
The benefits also ripple beyond you and your staffer. According to Cody Austin, CPA (Canada), an entrepreneur who wears several hats, including vice president of operations for Integrity Audit and Accounting Ltd., Calgary, Alberta, Canada, "Having great lines of communication creates a better, stronger team."
Unfortunately, not every manager receives training in how to conduct one-on-one meetings. Here’s how to get started:
Make it recurring. Having an occasional meeting won’t create the momentum needed for relationship building and guidance, so experts advise having weekly or biweekly meetings. Maureen Crawford Hentz, a Boston-based career coach and human resources professional, uses Wednesdays for meetings. Austin, meanwhile, blocks off 30 minutes on a day that works for both him and the employee. If the entire time isn’t needed, that’s fine, but "there’s always something I can talk about with them for 15 minutes," he said.
Consider the environment. While many meetings are being conducted virtually, when it comes to the office, the setting matters, Almonte said. If the goal is an informal catch-up, consider a walking meeting or a conference room. For serious meetings, Almonte avoids sitting behind a desk to avoid a position of power. "Remove barriers in the room to create an open space for the other individual," he said.
Set an agenda. Be sure that both sides know what will be brought up in the meeting in advance, giving everyone a chance to prepare responses. Crawford Hentz suggested including an agenda — it doesn’t need to be extremely detailed — in the meeting invitation. "Maybe you are an introverted thinker, or maybe your direct report is. Introverted thinking means that you need some time to process," she said. "You like to think about your reaction to something."
For Almonte, "a meeting without an agenda is like sailing without any directions."
Ask the right questions. If the employee is the type to open up, spend the first few minutes asking about how life is going outside of work, which will show that you’re invested in the whole person. "I need to know what my team is doing. We might spend five minutes finding out what’s going on — if their dog is sick, for example," Austin said. "We bring a lot of the things that affect us in our lives to work."
Then, transition into work-related inquiries. For Austin, examples include: "What’s on your plate? How is it going? What roadblocks are you encountering?" Crawford Hentz likes to ask: "What part of your job are you loving?" and "What part is your least favorite part?"
Don’t cancel. The manager should try not to cancel, Crawford Hentz said. If the direct report needs to cancel, he or she should check in before not attending the meeting, which keeps the line of communication open. "That way if you [the manager] have something you want to talk about, you can say, ‘Let’s just talk about this, and then we’ll leave the other things for next week/month.’ Or you can just say, ‘I’m here if you need me.’"
End it well. Wind down the meeting by discussing the next meeting and any potential schedule conflicts, Crawford Hentz said. Then consider ending on a positive note. You don’t want to pass out praise that isn’t deserved, but employees need to be reminded that they — and their efforts — are appreciated. When warranted, "end with an observation of their work this week or an appreciation you have for them," she said.
Follow up. Always be sure to follow up on the topics discussed, Almonte said. "Ensure that the message was received, action steps are being followed, and allow for the opportunity for feedback and questions," he advised.
Practice what you preach. Be sure that you’re following this advice with your leadership. "Communicating up the chain is just as important to my boss," Austin said. They discuss what’s going on with Austin and what he’s learning from direct reports, he said.
Start with these simple tips to get comfortable with one-on-one meetings. Managers who learn to successfully conduct one-on-one meetings, and practice what they preach, benefit themselves and their direct reports.
Caitlin Morin, CPA, a senior accountant who reports to Almonte, acknowledged getting nervous just like him. But she was quickly put at ease and gained valuable guidance.
"I not only left feeling that my ideas had been heard but also came away with concrete ways to develop my own skills, specifically when it comes to getting outside my comfort zone," she recalled. "These [weekly] meetings, as well as other one-on-one meetings, have been instrumental in my career growth and in building relationships with my higher-ups."
— Dawn Wotapka is a freelance writer based in Georgia. To comment on this article or to suggest an idea for another article, contact Courtney Vien at [email protected].
The Journal of Accountancy is now completely digital.
SPONSORED REPORT
Employees working remotely have created numerous issues for employers. The 2022 Payroll Update report provides insight on remote workforce tax issues, pandemic payroll issues and employer credits, and worker classification issues in the gig economy.
NEWS APP
This quick guide walks you through the process of adding the Journal of Accountancy as a favorite news source in the News app from Apple.
© Association of International Certified Professional Accountants. All rights reserved.
Reliable. Resourceful. Respected.
Dow, S&P Live Updates for Aug. 28, 2022 – Bloomberg
Bloomberg Markets Asia. Live from Hong Kong, bringing you the most important global business and breaking markets news information as it happens.
Bloomberg Law speaks with prominent attorneys and legal scholars, analyzing major legal issues and cases in the news. The show examines all aspects of the legal profession, from intellectual property to criminal law, from bankruptcy to securities law, drawing on the deep research tools of BloombergLaw.com and BloombergBNA.com. Reporters from Bloomberg’s Washington, D.C. bureau are prominently featured as they offer analysis of policy and legal issues.
The only daily news program focused exclusively on technology, innovation and the future of business from San Francisco.
China Arrests Hundreds in Nation’s Biggest-Ever Bank Fraud Probe
Yen’s Slump Toward 140 Reawakens Talk of Currency Intervention
Ambani’s Reliance to Spend $25 Billion on 5G Across India
SolarEdge Faces Potential Imports Ban as ITC Launches Probe
Peloton Delays 10-K Report to Sort Out Accounting Tied to Restructuring Efforts
Biden Administration Readies $1.1 Billion Arms Sale to Taiwan
Secret Service Agent for Trump Retires After Jan. 6 Scrutiny
Adani Becomes World’s Third-Richest Person as Wealth Surges
NBA Star Giannis Antetokounmpo Is Lending His Name to an ETF
N.J. Man Who Sold Fake Tom Brady Championship Rings Gets 3 Years in Prison
Movie Theaters Seek to Fill Seats With $3 Tickets This Weekend
Electric-Vehicle Leader BYD Has a Battery-Cost Problem
It’s Meatless Monday. Maybe Avoid the Salmon.
Neel Kashkari Says the Quiet Part Out Loud on Markets
The Search for New Talent Has Companies Welcoming Refugees
Crypto’s Real Value Was Never $3 Trillion
DraftKings Is Coming for Your Dumb Money at Wrigley Field
Arizona GOP Senate Candidate Suggests Fed Diversity to Blame for Economic Woes
US Freight Railroads Reach Tentative Deals With Three Unions
Former Top Economist at US Labor Department Joins SEIU Union
EXPLAINER: Pakistan Fatal Flooding Has Hallmarks of Warming
Fracking Titan Liberty Says Fossil Fuels to Rule for Decades
Surprise Visit to NYC’s Rikers Island Jail Finds Improvements
Over 1 in 10 New Texas Residents Migrated There From California
N.J. Gasoline Tax to Drop 1 Cent on Better-Than-Forecast Demand
Why Are Luxury Brands So Bullish on NFTs?
Singapore Mulls Making It Harder for People to Trade Crypto
Divide Emerges Between Retail, Pros on How to Play Ether Upgrade
Isabelle Lee
US stocks and Treasuries fell again Monday as the realization that interest rates are likely to remain elevated for an extended period continued to force a repricing across assets.
The S&P 500 and the Nasdaq 100 dropped a second day, adding to the rout that started Friday when Jerome Powell made it clear the Fed is willing to let the economy suffer as it fights inflation. Treasury yields rose, with the 10-year rate hovering around 3.11%. The two-year yield had climbed to its highest level since 2007 earlier in the day before paring the advance. Oil notched gains on supply risks.
Arts Centre Funding – The Journal of Music

The purpose of Arts Centre Funding is to invest in and support the infrastructure of arts centres required to sustain and develop the arts in Ireland.
In offering Arts Centre Funding, the Arts Council wishes to ensure:
– The year-round production and dissemination of arts activities, events and services of the highest quality throughout the country
– That greater numbers of people enjoy high-quality arts experiences
– That artists are supported in their professional practice
– That audiences, artists and participants reflect the diversity of contemporary Ireland
– Increased depth of engagement by and with the public
– High standards in governance and management.
See link below for further details.
Please note that some listings are added by third parties. The Journal of Music does not take responsibility for the content or accuracy of listings published by third parties on this site. The Journal of Music reserves the right to edit or delete listings. Click here to add a listing, login or register.
Tennessee judge must turn trial recordings over to convicted killer – Tennessee Lookout – Tennessee Lookout
(Getty Images)
A state appellate court is ordering an East Tennessee judge to turn over trial audio recordings he’s kept locked away in his office in a bid to keep the convicted killer he sent to prison from obtaining a copy.
In a first of its kind ruling, the Tennessee Court of Appeals says in a decision made public this week that audio recordings of criminal trials are public records under state law, and 8th Judicial District Criminal Court Judge Shayne Sexton should not have withheld from convicted killer Kevin Lee Waggoner a copy of audio from Waggoner’s trial.
“The record on appeal establishes that Judge Sexton currently has possession of the only copy of the audio recordings,” Appellate Judge Kristi M. Davis wrote in the opinion. “In congruence with (state law), the recordings must be immediately filed with the criminal court clerk for storage.
“For more than a century, Tennessee courts have recognized the public’s right to inspect governmental records,” Davis wrote. “This right extends to court records … The (Tennessee Public Records Act) governs access to public records and is intended to provide a tool to hold government officials and agencies accountable to the citizens of Tennessee through oversight in government activities.
“There can be no doubt, then, that the audio recordings are public records for purposes of the act,” the opinion stated.
Sexton, who is retiring at the end of this month, did not respond to a request for comment. It is unknown if he routinely keeps criminal trial audio recordings in his office or only did so in Waggoner’s case.
All criminal trials in Tennessee are recorded by a court reporter, who is supposed to file those recordings and a transcript made from those recordings with the court clerk in each respective county in which a trial is held.
Waggoner, a former Knox County school security officer convicted in the 2013 shooting death of his neighbor in Union County, has been fighting from behind bars for a copy of the audio recordings from his trial.
Waggoner, who is serving an 18-year prison term for second-degree murder, represented himself in the public records act lawsuit against Sexton and Union County Criminal Court Clerk Barbara Williams. Waggoner has argued he believes there are “discrepancies” between what occurred at his 2016 trial and what the written transcript of the trial shows, and that he needs a copy of the original audio recordings to determine if his belief is true.
Judge: Convicted killer might alter recordings
Waggoner was working as a security officer at Shannondale Elementary School when a nearly three-year feud with neighbor Michael Woodby, 45, ended with Woodby fatally shot four times, his body lying in a ditch several yards off state Highway 370 in Union County.
Just three days before the fatal shooting, Waggoner sent copies of video surveillance he and his family had been conducting to document their grievances against Woodby to local media outlets. Trial testimony would later show the Waggoner family had a history of feuds with neighbors, each characterized by video surveillance by the entire Waggoner family and angry confrontations.
The Waggoners videoed every bout in their feud with Woodby — except the shooting — and even recorded audio of police scanner traffic indicating that Woodby’s body was on its way to the morgue.
Waggoner contended he only shot Woodby after he attacked his adult son with a stick Woodby used as a cane. State prosecutors contended Waggoner ambushed Woodby when Woodby went for his nightly walk alongside the road that separated the two men’s homes.
Waggoner’s first trial in 2015 ended in a hung jury. He was convicted after a second trial a year later. Sexton sentenced him to 18 years in prison.
After Waggoner was convicted, he first sought a copy of the audio recording of his trial from Sexton during a motion hearing. Sexton refused the request. Waggoner then began filing requests for the audio via the Tennessee Public Records Act and eventually sued Sexton and Williams in Davidson County Chancery Court.
As part of that litigation, Sexton filed an affidavit saying trial audio recordings are not “part of the official record of Mr. Waggoner’s trial proceedings” and that he was keeping the Waggoner trial audio in his office because he feared Waggoner would tamper with it.
“The proof at trial established that (Waggoner) and his family were technophiles, very adept and skilled with electronic recording equipment both video and audio, and, in fact, (Waggoner) recorded his own 911 call to the police after killing the victim,” Sexton wrote in the affidavit.
“In addition, during the course of (Waggoner’s) trial, other members of his family engaged in conduct intended to disrupt the proceedings, including attempts on their part to record the court proceedings via a cell phone and/or transmit the court proceedings via cell phone over the internet,” Sexton continued.
“I prohibited these efforts,” the judge wrote. “It was, and continues to be, a concern of the court that if (Waggoner) is provided the audio recording of the proceedings or even a copy of the audio recording, he or members of his family will attempt to alter the recording in some fashion to create the appearance of an inaccuracy in the official trial transcript. To reiterate, the audio recording, which is in the possession of my office, is not part of the official record of Mr. Waggoner’s trial proceedings.”
Court: Attorney General contradicts own opinion
Davidson County Chancellor Patricia Head Moskal sided with Sexton. Waggoner appealed. In its ruling this week, the appellate court rejected Sexton’s arguments and struck down Moskal’s ruling.
“Judge Sexton in his sworn statement expressed concern that (Waggoner) could alter the audio recordings,” the court stated. “Nonetheless, the same could be said for any electronic subject to disclosure, and this argument is not grounded in any legal authority.
“While (Sexton) argued in the trial court that (Waggoner) might alter the audio recording, this is not a firm basis for denial of an otherwise public record inasmuch as the same could be said for nearly any public record,” the court stated. “Nor is the criminal court clerk required to provide (Waggoner) with the only copy of the recordings.”
The Tennessee Attorney General’s office represented Sexton in the litigation and argued on his behalf that trial audio recordings are not public record and do not have to be retained as public records in the clerk’s office.
But the appellate court pointed out the AG’s office was contradicting its own previously-published opinion.
“The Attorney General of Tennessee (in a 2008 advisory opinion) … concluded that if a court reporter utilizes audio recording devices or stenography to make verbatim recordings of a proceeding in a criminal case, the reporter must file the resulting recordings or stenographic notes with the court clerk, who shall preserve them as part of the records of the trial,” the appellate court stated.
“We are particularly persuaded by the (2008) opinion of the Attorney General in (Waggoner’s) case because it squarely answers the question before us, no appellate court of Tennessee has concluded otherwise, and the (2008) opinion has been held out to the courts of this state as the authority on (state law,)” the appellate court continued.
by Jamie Satterfield, Tennessee Lookout
August 29, 2022
by Jamie Satterfield, Tennessee Lookout
August 29, 2022
A state appellate court is ordering an East Tennessee judge to turn over trial audio recordings he’s kept locked away in his office in a bid to keep the convicted killer he sent to prison from obtaining a copy.
In a first of its kind ruling, the Tennessee Court of Appeals says in a decision made public this week that audio recordings of criminal trials are public records under state law, and 8th Judicial District Criminal Court Judge Shayne Sexton should not have withheld from convicted killer Kevin Lee Waggoner a copy of audio from Waggoner’s trial.
“The record on appeal establishes that Judge Sexton currently has possession of the only copy of the audio recordings,” Appellate Judge Kristi M. Davis wrote in the opinion. “In congruence with (state law), the recordings must be immediately filed with the criminal court clerk for storage.
“For more than a century, Tennessee courts have recognized the public’s right to inspect governmental records,” Davis wrote. “This right extends to court records … The (Tennessee Public Records Act) governs access to public records and is intended to provide a tool to hold government officials and agencies accountable to the citizens of Tennessee through oversight in government activities.
“There can be no doubt, then, that the audio recordings are public records for purposes of the act,” the opinion stated.
Sexton, who is retiring at the end of this month, did not respond to a request for comment. It is unknown if he routinely keeps criminal trial audio recordings in his office or only did so in Waggoner’s case.
All criminal trials in Tennessee are recorded by a court reporter, who is supposed to file those recordings and a transcript made from those recordings with the court clerk in each respective county in which a trial is held.
Waggoner, a former Knox County school security officer convicted in the 2013 shooting death of his neighbor in Union County, has been fighting from behind bars for a copy of the audio recordings from his trial.
Waggoner, who is serving an 18-year prison term for second-degree murder, represented himself in the public records act lawsuit against Sexton and Union County Criminal Court Clerk Barbara Williams. Waggoner has argued he believes there are “discrepancies” between what occurred at his 2016 trial and what the written transcript of the trial shows, and that he needs a copy of the original audio recordings to determine if his belief is true.
Judge: Convicted killer might alter recordings
Waggoner was working as a security officer at Shannondale Elementary School when a nearly three-year feud with neighbor Michael Woodby, 45, ended with Woodby fatally shot four times, his body lying in a ditch several yards off state Highway 370 in Union County.
Just three days before the fatal shooting, Waggoner sent copies of video surveillance he and his family had been conducting to document their grievances against Woodby to local media outlets. Trial testimony would later show the Waggoner family had a history of feuds with neighbors, each characterized by video surveillance by the entire Waggoner family and angry confrontations.
The Waggoners videoed every bout in their feud with Woodby — except the shooting — and even recorded audio of police scanner traffic indicating that Woodby’s body was on its way to the morgue.
Waggoner contended he only shot Woodby after he attacked his adult son with a stick Woodby used as a cane. State prosecutors contended Waggoner ambushed Woodby when Woodby went for his nightly walk alongside the road that separated the two men’s homes.
Waggoner’s first trial in 2015 ended in a hung jury. He was convicted after a second trial a year later. Sexton sentenced him to 18 years in prison.
After Waggoner was convicted, he first sought a copy of the audio recording of his trial from Sexton during a motion hearing. Sexton refused the request. Waggoner then began filing requests for the audio via the Tennessee Public Records Act and eventually sued Sexton and Williams in Davidson County Chancery Court.
As part of that litigation, Sexton filed an affidavit saying trial audio recordings are not “part of the official record of Mr. Waggoner’s trial proceedings” and that he was keeping the Waggoner trial audio in his office because he feared Waggoner would tamper with it.
“The proof at trial established that (Waggoner) and his family were technophiles, very adept and skilled with electronic recording equipment both video and audio, and, in fact, (Waggoner) recorded his own 911 call to the police after killing the victim,” Sexton wrote in the affidavit.
“In addition, during the course of (Waggoner’s) trial, other members of his family engaged in conduct intended to disrupt the proceedings, including attempts on their part to record the court proceedings via a cell phone and/or transmit the court proceedings via cell phone over the internet,” Sexton continued.
“I prohibited these efforts,” the judge wrote. “It was, and continues to be, a concern of the court that if (Waggoner) is provided the audio recording of the proceedings or even a copy of the audio recording, he or members of his family will attempt to alter the recording in some fashion to create the appearance of an inaccuracy in the official trial transcript. To reiterate, the audio recording, which is in the possession of my office, is not part of the official record of Mr. Waggoner’s trial proceedings.”
Court: Attorney General contradicts own opinion
Davidson County Chancellor Patricia Head Moskal sided with Sexton. Waggoner appealed. In its ruling this week, the appellate court rejected Sexton’s arguments and struck down Moskal’s ruling.
“Judge Sexton in his sworn statement expressed concern that (Waggoner) could alter the audio recordings,” the court stated. “Nonetheless, the same could be said for any electronic subject to disclosure, and this argument is not grounded in any legal authority.
“While (Sexton) argued in the trial court that (Waggoner) might alter the audio recording, this is not a firm basis for denial of an otherwise public record inasmuch as the same could be said for nearly any public record,” the court stated. “Nor is the criminal court clerk required to provide (Waggoner) with the only copy of the recordings.”
The Tennessee Attorney General’s office represented Sexton in the litigation and argued on his behalf that trial audio recordings are not public record and do not have to be retained as public records in the clerk’s office.
But the appellate court pointed out the AG’s office was contradicting its own previously-published opinion.
“The Attorney General of Tennessee (in a 2008 advisory opinion) … concluded that if a court reporter utilizes audio recording devices or stenography to make verbatim recordings of a proceeding in a criminal case, the reporter must file the resulting recordings or stenographic notes with the court clerk, who shall preserve them as part of the records of the trial,” the appellate court stated.
“We are particularly persuaded by the (2008) opinion of the Attorney General in (Waggoner’s) case because it squarely answers the question before us, no appellate court of Tennessee has concluded otherwise, and the (2008) opinion has been held out to the courts of this state as the authority on (state law,)” the appellate court continued.
Tennessee Lookout is part of States Newsroom, a network of news bureaus supported by grants and a coalition of donors as a 501c(3) public charity. Tennessee Lookout maintains editorial independence. Contact Editor Holly McCall for questions: [email protected]. Follow Tennessee Lookout on Facebook and Twitter.
Our stories may be republished online or in print under Creative Commons license CC BY-NC-ND 4.0. We ask that you edit only for style or to shorten, provide proper attribution and link to our web site. Please see our republishing guidelines for use of photos and graphics.
Jamie Satterfield is an investigative journalist with more than 33 years of experience, specializing in legal affairs, policing, public corruption, environmental crime and civil rights violations. Her journalism has been honored as some of the best in the nation, earning recognition from the Scripps Howard Foundation, the Society of Professional Journalists’ Sigma Delta Chi Awards, the Green Eyeshade Awards, the Tennessee Press Association, the Tennessee Managing Editors Association, the First Amendment Center and many other industry organizations. Her work has led to criminal charges against wrongdoers, changes in state law and citations in legal opinions and journals. She was married to the love of her life for 28 years and is now a widow and proud mother of two successful children of good character and work ethic.
DEMOCRACY TOOLKIT
© Tennessee Lookout, 2022
Now more than ever, tough and fair journalism is important. The Tennessee Lookout is your watchdog, telling the stories of politics and policy that affect the people of the Volunteer State.
Our stories may be republished online or in print under Creative Commons license CC BY-NC-ND 4.0. We ask that you edit only for style or to shorten, provide proper attribution and link to our web site.
VMAs Live Stream: Watch MTV's Official 2022 Red Carpet Pre-Show – TVLine
By
MTV is kicking off the 2022 VMAs a little early on Sunday with a star-studded pre-show that includes performances from several of this year’s nominated artists.
Best New Artist nominee Dove Cameron, MTV PUSH artist Saucy Santana and Yung Gravy are all set to perform during the 90-minute event, which also features DJ Murda Beatz as the official Kraft Singles House DJ.
The pre-show is hosted by Nessa and Kevan Kenney, with Tate McRae pitching in as a special celebrity correspondent. Watch it all go down live when the stream begins at 6:15/5:15c.
Doja Cat, Jack Harlow and Harry Styles enter Sunday’s ceremony at the head of the pack with eight nominations apiece, followed by Drake, Kendrick Lamar and Lil Nas X with seven. Billie Eilish is up next with six, while BTS, Dua Lipa, Ed Sheeran, Future, Lizzo, Taylor Swift and The Weeknd are all five-time nominees.
As always, the main show is where it’s at. Beginning at 8 pm, music enthusiasts can expect performances from Anitta, Bad Bunny, Blackpink, Eminem, J Balvin, Harlow, Kane Brown, Khalid, Lizzo, Måneskin, Marshmello, Nicki Minaj, Panic! at the Disco, Red Hot Chili Peppers and Snoop Dogg.
Hit PLAY on the video above to watch this year’s VMA pre-show, then drop a comment with your thoughts below. Which artists you rooting for?
Ralph Eggleston Dies: Longtime Disney & Pixar Animator, ‘For The…
MLBPA Recruits Minor Leaguers as Baseball and Its Players…
Michael Jackson Begged Warner Bros. to Star in Neil Gaiman’s…
Jimmy Fallon, Miley Cyrus and Willie Nelson Join ‘Dolly Parton’s…
The Most Corntastic Gifts for The Corn-Obsessed
TVLine is a part of Penske Media Corporation. © 2022 TVLine Media, LLC. All Rights Reserved.
PMC Entertainment
Powered by WordPress VIP
Online Master of Agriculture | WTAMU – West Texas A&M University
Take the next step to create a brighter future with an online Ag degree from WTAMU. This exciting new degree can help you achieve your goals in the field of Agriculture. We’ve taken the time and put in the effort to map out a path to real-world readiness. We promise you more than a general online degree, we promise you the Buff Experience. An experience you’ll never forget – faculty advising, career development and more.
We offer variability in our teaching, to give you both one-on-one time with our faculty with enough flexibility to learn when you want and where you want. Our courses are led by expert faculty who aim to give you the same learning as their on-campus courses.
Our faculty are passionate about the field of Agriculture in a variety of areas from Education to Agronomy. More importantly, they’re passionate about students and creating an open, academically rigorous learning environment. You’ll work with top educators who will support you through your degree program from the minute you log on to your graduation.
Entrepreneur set on making a difference in housing industry – Business Observer
The Observer Media Group Inc. respects your privacy and values the relationship we have with you. We use technology to collect information that helps us enhance your experience and our products and services. The cookies we use help us understand what information and advertising are most useful and valuable. Please take a moment to review our Cookie Policy and Privacy Policy. By clicking “Accept,” you agree to our use of cookies and similar technologies.
While working at a dream job in California, Vengroff got a call from his father that would change his trajectory and lead him to Florida with a passion to provide affordable housing for an underserved community.
About eight years ago, Mark Vengroff, 56, found himself as CEO of his dream job.
At the time, he was working with the founder of Clearlight Partners, an advisory private equity firm that had recently purchased Walker Advertising. Soon enough, Vengroff was named CEO of Walker Advertising, which assists legal firms with marketing and branding, including filming commercials.
But two years into that job, his father, the late Harvey Vengroff, a prominent Sarasota businessman known partially for his battles with local officials over zoning and housing regulations, called with some news. The elder Vengroff had acquired a third hotel property and shortly after was diagnosed with cancer. Given a year to live, Harvey asked his son to take over managing the properties. Resisting at first, the younger Vengroff asked his father to “just sell them. I was happy doing what I was doing.”
“He said, ‘I can’t sell it. Thousands of families will be without a place to live,’” Vengroff says. Harvey had purchased the first hotel building 20 years ago in Sarasota and rented it out to families. When he made the call to Mark, he had a second hotel property in Sarasota and one in Bradenton that was under construction.
So, two years into his time at Walker Advertising, Mark Vengroff flew from California to Florida to get a sense of the situation. After seeing the communities the residents had formed, he knew he couldn’t let them, or his father, down.
Not wanting to leave Walker Advertising without a CEO, he stayed for six months, until a new leader was in place. Then he spent the last six months of his father’s life working “side by side” with him to grow the company his father created: One Stop Housing. The Sarasota-based company specializes in buying and converting hotels, offices, schools and other property types into apartment buildings whose units are priced below market rates. It operates in Bradenton, Kissimmee, Orlando, Sarasota and Memphis, Tennessee. (Harvey Vengroff died Oct. 11, 2018. He was 77.)
“We’re for-profit, but have a philanthropic approach,” Mark Vengroff says. “Everyone deserves a place to live.”
Five years later, the business has grown substantially. Currently, there are 800 units in the queue to be built this year. That’s in addition to the 3,000 units already been built, plus the eight hotels it's converted to workforce housing.
The company keeps rates within the 60-80% of the Area Median Income (AMI) range, but in order to live within a One Stop Housing community, prospective residents must have a job. That's the only requirement.
The average cost to live in a One Stop Housing community is $750-800 a month, which is all-inclusive with utilities. The goal? For workers to live close to where they work without spending their entire paycheck on housing — a big issue in hospitality-centric communities like Sarasota or Orlando.
Vengroff can charge more, especially in today's lava-hot apartment rental environment.
“We’re 40% below market,” Vengroff adds. “My father said, ‘greed kills everything good.’ If we’re making enough, we don’t have to take it out on the tenant.”
When looking for a property to build a community, One Stop Housing has few limits. Vengroff says the company has acquired everything from schools to commercial and industrial spaces. As long as the structure, or land, is near public transportation, it can work.
After One Stop Housing acquires a property, it’s time to either renovate or build. Vengroff says the company does so without cutting corners, even though construction and material costs have recently “gone through the roof.” The company cuts down on a lot of those costs by using its in-house construction team.
Since the layouts for the apartments are similar, One Stop Housing can purchase in bulk. So even though the cabinets are custom-made, the company could purchase 800 at a time to reduce cost. The four apartment layouts include a studio, one bedroom, two bedroom with one bathroom and two bedroom with two bathrooms.
The biggest motivator for the team and Vengroff to this day remains Harvey. “My father was always my hero. His spirit lives very loud and clear here,” he says, noting especially of Harvey’s philanthropic focus. “We keep that in the back of our heads.”
“His final words to me were, ‘grow this thing to a billion,’” which was a nod to earlier in Vengroff’s career. As CEO of Vengroff, Williams and Associates, a debt collection agency started by Harvey, Vengroff grew the company into 19 countries with 1,100 employees and took it from managing $10 million to over $32 billion in assets.
Of his own passion for providing workforce housing, that’s an easy question for Mark Vengroff to answer. The tenants and their stories.
“It always seems when I have a bad day, I’ll get a phone call or letter from a tenant,” he says.
One story Vengroff tells is particularly moving.
A woman, who left a bad situation in Texas, moved in with her friend in Florida. Not long after, she found herself kicked out of the temporary housing. She was widowed, had no credit and was living on the street, even though she had a job, Vengroff says.
Every housing place she tried denied her application. Following a particular rough day of being turned away, she devised a plan. There was one more place on her list to apply for housing. If that place rejected her, she planned to go to the Sunshine Skyway Bridge that spans Tampa Bay between Hillsborough and Manatee counties to jump, and kill herself. That last place was One Stop Housing.
After a brief interview, she was handed an application to get her set up for an apartment. In disbelief, tears started streaming down her face. Vengroff says when asked why she was crying, she says, “you just saved my life.”
“This," Vengroff says, "is exactly why I’m doing what I’m doing.”
The One Stop Housing company truly is one stop.
Mark Vengroff kept on the managerial team that was under his father and then acquired a construction company to source all the company’s construction needs in-house. He also started a nonprofit, One Stop Cares, to assist with all of the “wrap-around” services, like providing residents with furniture, food or even help with an interview, and appointed a president to oversee operations. The nonprofit has also hosted toy drives, financial services classes, educational programs, parenting 101 classes and even provided a place for residents to get COVID-19 shots.
The company has also opened a telehealth center so residents have access to healthcare in their clubhouse, and Vengroff says they’re looking to offer this capability throughout more of their communities.
It's even gone as far as purchasing a $100,000 home, fixing it up for about $50,000 and handing the tenant the title. They set the tenant up on a mortgage plan to help repair their credit and within two years the tenants have paid back the $150,000. The company has only done this about a dozen times.
One Stop Housing is also home to Florida Incubator, a division of the nonprofit. After realizing there were tenants who were running businesses out of their homes, Vengroff created the division to align tenants with a coach before they speak with investors. “It’s really about the coaching and helping the entrepreneur,” he says.
So far, Vengroff's seen tree-cutting, pool and lawn services, and T-shirt companies thrive through the program, as well as many others including technology-focused companies.
Amanda Postma is a business reporter covering Sarasota and Manatee Counties, a position she’s held since 2021. After graduating from the University of Missouri School of Journalism in 2018, Amanda was a reporter for a small-town newspaper in Missouri before becoming a marketing…
View the discussion thread.
USPS Periodicals Permit
ISSN# 2325-8195
as required for Public Notices per F.S. 50.011
Corps seeks public input on revised Coralville Lake Reservoir Master Plan – mvr.usace.army.mil
Aerial view of Coralville Lake dam and reservoir.
The U.S. Army Corps of Engineers, Rock Island District is finalizing the Master Plan for Coralville Lake Reservoir and is seeking public input. A digital version of the plan and a video outlining changes are available for review at: https://www.mvr.usace.army.mil/Missions/Recreation/Coralville-Lake/Master-Plan/.
A virtual open house will be hosted Wednesday, Sept. 7, at 5 p.m. to offer a question-and-answer opportunity for the public and to gather feedback about the updated plan. Details on how to submit comments and participate in the virtual public open house are also available at the website listed above.
Master Plans are a vital tool used to guide responsible stewardship and management of public lands administered by the U.S. Army Corps of Engineers. These documents provide direction for appropriate land management regarding the use, development, enhancement, protection, and conservation of the natural, cultural and man-made resources at USACE civil works projects throughout the country.
For more information on the plan revision process, the updated plan or how to submit comments, visit: https://www.mvr.usace.army.mil/Missions/Recreation/Coralville-Lake/Master-Plan/. Comments will be accepted through September 15, 2022 and can also be submitted by email to: [email protected], telephone to: 309-794-5704, or by mail to: Rock Island District, Attn: PM-M, P.O. Box 2004, Rock Island, IL 61204-2004.
Release no. 22-022
Deliver vital engineering solutions, in collaboration with our partners, to secure our Nation, energize our economy, and reduce disaster risk.
The official public website of the Rock Island District, U.S. Army Corps of Engineers. For website corrections, write to [email protected]
Efficient liquid phase confiscation of nile blue using a novel hybrid nanocomposite synthesized from guar gum-polyacrylamide and erbium oxide | Scientific Reports – Nature.com
Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Scientific Reports volume 12, Article number: 14656 (2022)
1
Metrics details
In recent times, biopolymer-metal oxide nanocomposites have gained prominent importance in the attenuation of environmental toxicants from aqueous phase. But lanthanide oxide-based biopolymer nanocomposites have scantly been evaluated for their adsorption potential. A novel guar gum-polyacrylamide/erbium oxide nanocomposite (GG-PAAm/Er2O3 NC) adsorbent was synthesized by copolymerization of guar gum (GG) and acrylamide (AAm) utilizing N–N′-methylenebisacrylamide as a crosslinker and Er2O3 as a reinforcing agent. The adsorptive efficacy of GG-PAAm/Er2O3 nanocomposite was evaluated using nile blue (NB) as a model pollutant dye from aquatic system. The prepared adsorbent was characterized by Fourier transform infrared (FTIR) spectroscopy, X-ray diffraction (XRD) analysis, Brunauer–Emmett–Teller (BET) analysis, thermogravimetric analysis, scanning electron microscopy-energy dispersive X-ray spectroscopy (SEM–EDX), and high-resolution transmission electron microscopy (HRTEM). The optimal process parameters, which include dosage (0.8 g/L), agitation time (40 min), initial solution pH (6), and initial NB concentration (80 mg/L) were determined by batch methodology. The equilibrium data for NB confiscation was better expressed by Langmuir isotherm model, with maximal adsorption effectiveness (Qm) of 225.88 mg NB/g demonstrating the actively monolayer adsorption onto homogeneous surface of GG-PAAm/Er2O3 NC. The kinetics of NB sorption process onto GG-PAAm/Er2O3 NC was reliable with pseudo-second order model. Thermodynamic parameters such as ΔH° (15–17 kJ/mol) and ΔS° (0.079–0.087 kJ/mol/K), and − ΔG° (8.81–10.55 kJ/mol) for NB validated the endothermic, an increased randomness at the GG-PAAm/Er2O3–NB interface, and spontaneity and feasibility of the process, respectively. The spent nanocomposite was effectively regenerated with NaOH, and could be reused proficiently for five runs demonstrating the high reusability potential of the nanocomposite. The commendable removal efficiency and high reusability of GG-PAAm/Er2O3 NC recommended it to be a highly competent adsorbent for cationic dyes particularly NB diminution from aqueous waste.
The control of environmental deterioration of aquatic resources resulting from the amplified discharge of liquid effluents due to rapid industrial development and global population growth has emerged as one of the major challenging tasks over the past several decades. Many industries such as textile, leather, tannery, cosmetics, paints or plastics are prime contributors of colored pollutants. The vast usage of synthetic dyes, which are mostly poorly-degradable and environmentally persistent due their complex molecular structures, in the textile industries emancipates huge quantity of dyes-laden water post dyeing and finishing operations. The release of unspent dyes in the receiving bodies not only worsens the water quality but presents a detrimental outcome including impaired photosynthesis of aquatic ecosystems, and carcinogenic, mutagenic or teratogenic influence on aquatic biota and humans1,2,3. Most dyes, on ingestion, are responsible for several diseases such as dizziness, vomiting, tremors, nausea, cyanosis and jaundice. Direct contact may result into allergic problems, skin irritation, and eye burn that can permanently damage the cornea. If inhaled, they may lead to breathing difficulty, profuse sweating, abdominal pain and hyper motility2. In the last few decades, much attention has been focused on mitigating the colored contaminants with a view to protecting the environmental sustainability and reduce the grave health risks associated with the industrial dyes4,5. Nile blue (NB) is an azo dye widely applied for dyeing in the textile industries. Many health-related problems including skin irritation, dermatitis, allergic reactions in eyes and respiratory disease are associated with its presence in water6. It can cause sleepiness, digestive system stimulation, cold feelings, mouth and throat irritation, redness and dryness of the skin, and chromosomal aberrations. Therefore, it is essential to decolorize the effluents containing the hazardous nile blue effectively before discharging into the aquatic system7.
The most common approaches for confiscating and decolorizing dyes from aquatic system are chemical coagulation, ozonolysis, photocatalysis, microbial degradation, electrochemical methods, and adsorption8,9,10. But most physico-chemical techniques are rather expensive, operationally cumbersome and energy-intensive. However, adsorption technology has attained widespread recognition as the most encouraging, reliable, highly effective, operationally simpler, efficient and cheap practice for environmental pollutants abatement with admirable regenerative and recyclable potential of the adsorbents.
Multitude of adsorbents such as activated carbons10, zeolites11, graphene oxides12, and clays13 have arduously been used for dyes decontamination. However, these adsorbents lack the traits of an ideal adsorbent, which has led to the pursuit for advanced adsorbents possessing higher sorption efficiencies. Over the recent years, biopolymers as adsorbents have gained popularity due to their environment-friendly characteristics14. However, due to their poor mechanical strength and fair water solubility, their usage in water pollution remediation is rather restricted. The utilization of metal oxides as filler in polymer matrices have minimized the strength related problem and has broadened their scope in water treatment. Metal oxides-based gum polysaccharide nanocomposites incorporating the synergistic assets of both the components have aroused a great deal of interest in aquatic environmental contaminants sequestration15,16,17 owing to their cheap and facile obtainability, feeble-toxicity, environmentally benign nature, improved surface area to volume ratio, easy modifiability, better biocompatibility, excellent mechanical strength, and magnificent adsorption effectiveness17,18,19,20. Many nanocomposite adsorbents derived from gum arabic21, gum karaya22, gum ghatti23, gum xanthan24, gum tragacanth25 have recently been described in the literature for sequestration of organic dyes from aquatic environment.
The modification of biopolymers is usually undertaken to impart certain functional groups/moieties to improve upon the surface characteristics and adsorptive performance. The nanocomposites based on guar gum (GG), a low-priced, innocuous, and biodegradable biopolymer, have shown promising applications as effective adsorbents for a number of organic and inorganic aquatic contaminants in the recent times26,27,28,29,30. However, to our best understanding, the adsorption studies on biopolymer/lanthanide oxide (nano)composites for the attenuation of diverse contaminants have hitherto received scant consideration. Recent researches on the removal of salicylhydroxamic acid and acid blue 92 by Nd2O331,32, rhodamine B by r-GO/rare earth metal oxide aerogels33, reactive black 5 using LaFeO3/chitosan nanocomposite34, arsenite by carrageenan-embedded LaFeO3 nanocomposite35, and metal ions by ceria and its composites36 have been undertaken. Additionally, grafted/cross-linked GG composites, Ag NPs containing GG/poly(AA) grafted polymer37, guar gum grafted acrylic acid/nanoclay38, mGG-PAAm39, GG/magnetite/chitosan40, pectin-crosslinked-GG/SPION41, GG-crosslinked-graphene oxide hydrogel42 and GG-metal oxide (nano)composites, ZnO NPs/GG43, Fe3O4-GG44, and TiO2/GG hydrogel45 have been arduously employed for the confiscation of many environmental pollutants. However, adsorption studies invoking guar gum cross-linked with PAAm (polyacrylamide) using N, N-methylenebisacrylamide as a cross-linker and erbium oxide as the filler has not been attempted so far to the best of our knowledge. Further, its utilization as an adsorbent for nile blue is also undocumented.
The present investigation aims at developing a novel and recyclable nanocomposite adsorbent for organic dyes remediation based on erbium oxide-reinforced guar gum-polyacrylamide biopolymer matrix and to evaluate its adsorption competence towards an organic cationic dye. The synthesis of guar gum-polyacrylamide/erbium oxide nanocomposite (GG-PAAm/Er2O3 NC), its characterization through infrared and nuclear magnetic resonance spectroscopy, X-ray diffraction, N2 adsorption/desorption, and thermogravimetric analyses, and scanning and transmission electron microscopy, and its liquid phase removal efficacy for nile blue (NB) is depicted in the current study. The process variables (agitation time, initial solution pH, dosage, and initial NB concentration) impacting the NB removal was explored to ascertain the optimal operating conditions, and various facets of the adsorptive uptake phenomena was explicated in terms of isotherm and kinetic parameters deduced by applying non-linear regression analyses of the adsorption data using the corresponding model equations. Thermodynamics parameters were also examined to estimate the energetic changes accompanying the adsorption process. Several adsorption–desorption cycles were performed in order to estimate the reuse potential of the adsorbent.
Guar gum (Loba chemie, India), Acrylamide (Spectrochem, India; 99.5%), N,N′-methylenebisacrylamide (MBA; Loba chemie, India; 99%), nile blue (Himedia, India; 96%), ammonium persulfate (APS) (Merck, India; 98%) were used as acquired without further purification. All chemicals were of AR standard. Double distilled water was employed for the preparation of working solutions. The operating pH of test solutions was adjusted between 2 and 10 by applying dilute HCl and NaOH solution (0.1 M). The FTIR spectra, XRD pattern, SEM images and TEM micrographs were obtained using Perkin-Elmer spectrometer, model BX spectrum, USA, Philips Analytica PW 1830 apparatus, (Philips, The Netherlands), Carl Zeiss JOEL scanning electron microscope, (Sigma 5.05, Germany) and Transmission electron microscope, HRTEM 200 kV model, FEI (Tecnai), respectively. TGA analyses (35 oC–600 oC) were performed using a Perkin-Elmer thermal analyser (TGA 4000, Massachusetts, USA, Pyris 6 TGA with PyrisTM software V. 11.0.0.0449) at a heating rate of 10 oC min–1 under nitrogen atmosphere. The initial and residual NB concentration was determined using a UV-visible spectrophotometer (T-80 + UV/Vis spectrophotometer, PG Instrument Ltd, UK).
The GG-PAAm/Er2O3 nanocomposite was prepared by cross-linking guar gum and acrylamide by employing MBA as a cross-linker in the presence of Er2O3 (filler). Aqueous suspension (100 ml) of guar gum (1.0 g), prepared by vigorous stirring for 1 h, and acrylamide solution (0.3 g in 10 mL of deionized water) were homogenously mixed under constant stirring at 60 °C followed by the addition of APS (1.0 g in 5 mL water) as initiator and Er2O3 as filler. N,N′-Methylenebisacrylamide (0.2 g) was added to the mixture under continuous stirring for 30 min, and the reaction mixture was sonicated for 1 h. The obtained product (%yield = 72) was washed with water and ethanol to remove unreacted moieties, oven-dried at 90 °C for 48 h, ground to a fine powder, and stored in a stoppered glass vial for further studies.
The impact of different operating factors such as dosage, agitation time, initial solution pH, initial NB concentration, and temperature on removal efficiency of GG-PAAm/Er2O3 nanocomposite adsorbent was evaluated by batch adsorption experiments, wherein NB solution (50 mL) of known concentration (20, 30, 40, 50 mg/L) was agitated in a water bath shaker with desired adsorbent mass (0.2–1.2 g/L) for different contact time (10–60 min) at constant temperature. After agitating for the desired time interval, the solution was centrifuged and the [NB] in the supernatant was determined spectrophotometrically at maximum absorbance wavelength of 635 nm. Adsorption isotherm studies at different temperatures (303, 308, and 313 K) was accomplished by shaking NB solution of various concentrations (30–80 mg/L) till equilibrium was achieved. However, for kinetics appraisal, similar experiments were conducted with optimal nanocomposite dosage (0.8 g/L) at 303 K and pH 6 using varying [NB] (30, 40 and 50 mg/L) for optimal agitation time (40 min).
The FTIR spectra of GG-PAAm/Er2O3 nanocomposite, prior and post adsorption, is shown in Fig. 1. The diminished intensity of peaks around 3000–3500 cm−1 for GG-PAAm/Er2O3 NC signified the interaction of hydroxyl groups of guar gum with amide group of polyacrylamide46. The peaks detected in the spectra around 1200 cm−1 depicted the C–C–O, C–OH and C–O–C stretching modes of polysaccharides14. The peaks at 1654 cm−1 and 1604 cm−1 were due to C=O stretching vibrations in acrylamide47. The peak observed at 1081 cm−1 was attributed to bending vibration for CH2–O–CH248, whereas the band at 1410 cm−1 due to C–N stretching vibrations49. The peak at 3028 cm−1 was ascribed to NH2 stretching vibrations of the polyacrylamide50. Additionally, the band at 659 and 608 cm−1 were attributed to Er–O–Er and Er–O linkage, respectively51, which clearly indicated the existence of metal–oxygen bond participating in the biopolymer nanocomposite. After NB adsorption (Fig. 1b), the FTIR spectrum showed slight shift in the peak position assigned to NH2 from 3028 to 3181 cm−1, and diminution of intensities together with slight shifting of absorption bands at 1654 and 1081 cm−1 to lower wavelengths indicating that the relevant functional groups were involved in the adsorption procedure either by van der Waals forces or hydrogen bonding9. Further, a slight shift in the O–H, C–O, and C–N bands; and a drop in the intensity of O–H stretching vibration at 3200 cm−1 after NB sorption, confirmed the interaction of dye molecules with existing functional groups10. Appearance of a new peak at 2934 cm−1 because of CH2 group of the aromatic ring structure of dye validated the sorption process onto the nanocomposite52.
FTIR spectra of (a) GG-PAAm/Er2O3 nanocomposite, and (b) NB-loaded nanocomposite.
The XRD diffraction pattern of Er2O3 is depicted in Fig. 2a that illustrated high crystallinity and sample purity. The main reflections of Er2O3 centred at 2ϴ = 19.56 (211), 28.60 (222), 32.82 (440), 42.72 (134), 48.54 (440), and 56.98 (622) matched completely with the JCPDS file no. 78–0390 having space group of Ia-353. The existence of characteristic peaks for erbium oxide in the diffraction pattern along with some shifting of the diffraction peaks and the formation of a semi-crystalline network due to the amalgamation of Er2O3 into the amorphous guar gum phase signified the fabrication of GG-PAAm/Er2O3 NC (Fig. 2b). Average crystallite size of GG-PAAm/Er2O3 NC was 57 nm.
XRD spectra of (a) Er2O3, (b) GG-PAAm/Er2O3 nanocomposite, (c) N2 adsorption–desorption curve, and (d) pore size distribution.
The N2 adsorption/desorption isotherm was used to determine the specific surface area and porosity of the nanocomposite. Higher surface area is related to greater accessibility of adsorptive sites, hence boosted adsorption aptitude of an adsorbent. The N2 adsorption/desorption isotherms of GG-PAAm/Er2O3 NC is shown in Fig. 2c,d. The specific surface area, pore volume and pore diameter of nanocomposite were 70 m2/g, 0.024 cm3/g and 5.796 nm, respectively. The pore diameter in the range of 2–50 nm confirmed the mesoporous nature of the prepared NC.
The thermal stability and rate of decomposition of the nanocomposite was gauged by TGA. This analysis could manifest the change in mass of the sample throughout the heating process. The TGA curves of GG and GG-PAAm/Er2O3 NC are shown in Fig. S1 revealing that the decomposition occurred in two stages. The first decomposition stage of GG started at 68.6 °C and ended at 168.9 °C with a loss of 10.13%, while in GG-PAAm/Er2O3 NC it started at 86.3 °C and terminated at 199.8 °C with a loss of 2.1%. The logical explanation for these results could be the moisture content loss associated with the gum. The second phase corresponding to the decomposition of sugars in GG started at 255.9 °C and terminated at 340.7 °C with a weight loss of 42%, while in GG-PAAm/Er2O3 this phase began at 264.7 °C and lasted till 342.9 °C with a weight loss of 38%. The results portrayed that the nanocomposite with Er2O3 as filler has acquired higher thermal stability with lesser mass loss than the parent gum.
The SEM micrographs along with related EDX spectra of GG-PAAm/Er2O3 NC and NB-loaded GG-PAAm/Er2O3 NC are presented in Fig. 3a,b. The morphology of GG-PAAm/Er2O3 NC displayed an irregular, uneven rough surface with heterogenous porous structures probably because of the cross-linking network suitable for adsorption. The existence of sufficient pores of different sizes and shapes were primarily accountable for greater surface area and admirable adsorption efficacy of GG-PAAm/Er2O3 NC. Figure 3b of GG-PAAm/Er2O3 NC, after confiscation of NB, exhibited almost smooth texture confirming the adsorption of NB onto GG-PAAm/Er2O3 NC. The EDX spectra for corresponding SEM micrographs are also revealed in Fig. 3a,b. The existence of C, N in the EDX spectra of NB-sorbed GG-PAAm/Er2O3 NC recommended the successful confiscation of NB onto the adsorbent surface. Considerable changes in the surface morphology of GG-PAAm/Er2O3 NC occurred after the sequestration of NB. The pores nearly disappeared, which might possibly be due to the occupation and entrapment of NB molecules in the pore structures.
SEM micrographs of (a) GG-PAAm/Er2O3 nanocomposite, and (b) NB-loaded nanocomposite along with corresponding EDX spectra.
To examine the structural morphology of GG-PAAm/Er2O3 NC, TEM investigation was performed. Figure 4a displays the TEM image for GG-PAAm/Er2O3 NC. The average size of GG-PAAm/Er2O3 NC was determined employing Image J software. The particle size distribution curve (Fig. 4b) suggested that the typical particles size ranged from 60 to 70 nm, that was in good agreement with XRD data. Moreover, the TEM image also confirmed the efficacious incorporation of Er2O3 within the biopolymer matrix. The grey portions in Fig. 5a demonstrated the GG and PAAm matrix, while the darker portions were accredited to Er2O3 nanoparticles randomly distributed into the GG-PAAm polymer matrix. In addition, such morphological features offered the GG-PAAm/Er2O3 NC a larger surface area.
TEM micrographs of (a) GG-PAAm/Er2O3 nanocomposite, and (b) histogram.
Effect of operational parameters on the removal of NB: (a) dose, (b) contact time, (c) initial concentration, and (d) initial solution pH.
The sorption characteristics of the nanocomposite for NB elimination was investigated in terms of different influencing parameters such as adsorbent dosage, initial NB concentration, agitation time, and initial solution pH at 303 K.
The reliance of aqueous phase removal of NB on variation in dosage was explored in 0.2–1.2 g/L range at pH 6, and the data is presented in Fig. 5a. The %R initially increased from 93.87 to 97.37 with an increment in adsorbent loading up to 0.8 g/L due to greater available active sites onto nanocomposite surface or an overall improved surface area, followed by a slow decrease thereafter. This decline in uptake efficacy on further increment in dosage might be due to higher accessible active sites relative to the number of NB molecules54. The removal capacity, however, showed an opposing trend with increase in adsorbent loading, which might be attributed to the overlapping of active sites, and/or decrease in surface area due to aggregation of adsorbent particles at higher dose that limited the removal capacity54,55. Therefore, 0.8 g/L of nanocomposite was selected as optimal dose for further studies. A similar trend of improved confiscation efficacy for NB with an increment in dosage of AC/CoFe2O4 nanocomposite56 and CuWO4 nanoparticles57 are described in the literature.
The agitation time is critical and significant factor in removing perilous pollutants. The effect of shaking time (10–60 min) on the sorbed NB onto the nanocomposite (0.8 g/L) is depicted in Fig. 5b. A rapid adsorption of NB occurred in the first 30 min (94.87–96.95%) suggesting a fast confiscation rate of NB molecules, gradually attaining equilibrium at 40 min with 97.62% removal. So, 40 min was preferred as the optimal equilibrium time for further studies. Initially, the removal process was rapid because of the existence of sufficient active surface-sites for absorbing NB. Subsequently, a decrement in the adsorption rate on increasing the shaking time was associated with the saturation of surface-active sites leading to decrease in the adsorption effectiveness. Similar trend in the NB removal was reported for clay/starch/MnFe2O458 and CNT/MgO/CuFe2O4 magnetic composite59 but with relatively higher equilibrium time of 60 min and 50 min, respectively.
The impact of change in the initial NB concentration (30–80 mg/L) on removal process was examined using optimal GG-PAAm/Er2O3 NC dosage (0.8 g/L) and time (40 min) at pH 6. The obtained results, depicted in Fig. 5c, revealed that the sequestration of NB declined from 97.08 to 95.93% with increasing initial NB concentration from 30 to 80 mg/L, which could be understood in terms of two adversative effects. A fixed mass of GG-PAAm/Er2O3 NC (0.8 g/L) has a definite number of surface-active sites. At low concentrations, the surface binding sites overwhelm the feeble number of NB molecules, which resulted in higher removal efficacy. However, when the solution concentration was increased further, the NB molecules progressively occupied the vacant sites, which reduced the number of available active sites. At high concentrations, fewer dye molecules occupied the remaining surface sites, hence decrease in removal efficiency. The increment in qe (36.40–95.93 mg/g) at higher solution concentration might be as a consequence of higher interactions between dye molecules and GG-PAAm/Er2O3 NC, or higher concentration gradient and/or increased driving force surpassing the mass transfer60. The optimal concentration for NB was 80 mg/L. Similar outcomes for NB adsorption with change in initial NB concentration were noticed for iron oxide nanoparticles61 and acrylamide- or 2-hydroxyethyl methacrylate-based copolymeric hydrogels62.
The solution pH is a crucial factor that plays a significant role in the removal process of contaminants. The extent of sorption is controlled by surface charge of the nanocomposite and the ionization of sorbate species that are governed by the solution pH63. Thus, the impact of the solution pH on NB confiscation was scrutinized in 2–10 pH range at optimum operating conditions (Fig. 5d). The pH of the solution reflects the nature of the physicochemical interactions of the NB molecules and the active sites of the GG-PAAm/Er2O3 NC. The pKa value of NB is 9.27 revealing its existence in the cationic form in the studied pH range. The pHzpc (= 5.6) of the nanocomposite meant that the surface was positively charged at pH < pHzpc implying an electrostatic repulsion between positive GG-PAAm/Er2O3 NC surface and cationic NB, which might have declined the removal rate in the pH range of 2–6. However, as the %uptake was significantly affected, it implied that other type of interactions like hydrogen bonding and π–π interactions might be responsible for higher removal rate (82.43–90.6%) in the 2–6 pH range. Further increase in the pH changed the surface of nanocomposite to negative, which made electrostatic attractions to be a part of removal mechanism and the %uptake augmented to 95.66% till pH 9. After that, the dye became negative and electrostatic repulsion declined the %removal. Similar trend was depicted for NB adsorption onto MoO3/Ppy nanocomposite6.
The analyses of isothermal equilibrium data at constant temperature by applying different isotherm models provide efficacious perspective with regard to maximum sorption capacity, homogeneity or heterogeneity of the adsorbent surface, affinity of the adsorbent towards adsorbate, coverage type, energy of adsorption, and the mechanism of adsorption. The corresponding isotherm parameters were determined using Ce versus qe plots employing Langmuir (Eq. 1), Freundlich (Eq. 2), Temkin (Eq. 3) and D-R (Eq. 4) models.
Langmuir isotherm asserts the sorption of NB onto the nanocomposite surface with finite number of energetically equivalent surface-active sites64 having equal affinity for NB molecules leading to monolayer formation.
Here, Ce (mg/L) and qe (mg/g) is the residual equilibrium NB concentration in fluid phase and the amount of NB sorbed onto the solid phase, respectively, Qm (mg/g) signifies maximum adsorption efficiency of the nanocomposite required to form monolayer of sorbate on its surface, b (L/mol) is a Langmuir constant, Kf (mg/g)(L/mg)(^{{1}/{text{n}_{text f}}}) is an indicator of sorption efficiency, nf indicates heterogeneity of nanocomposite surface and mutual interaction between sorbed species, 1/nf signifies functional strength of sorption, Kt (L/g) is binding constant associated with maximal binding energy, βt (= RT/bt) is a constant related to heat of adsorption, and qD (mg/g) is D–R sorption effectiveness.
The dimensionless factor, RL = (frac{1}{1+b{C}_{e}}) is used to evaluate the feasibility and favorability of adsorption procedure.
The corresponding isotherm model parameters along with correlation coefficients (R2) and standard error of estimate (SEE) at the studied temperatures were estimated from qe versus Ce curves for Langmuir (Fig. 6a), Freundlich (Fig. 6b), Temkin (Fig. 6c) and D-R (Fig. 6d) isotherms, and are tabulated in Table 1. An increment in the computed Qm values from 195.16 to 225.88 mg/g with an increase in operating solution temperature (303–313 K) (Table 1) indicated an improvement in the adsorption aptitude of the nanocomposite at higher temperature probably resulting due to enhanced physical attachment between active binding sites and NB molecules, which designated the removal process as endothermic. The b parameter (0.061–0.069 L/g) varied in the order: 303 K < 308 K < 313 K, which accounted for the best NB-nanocomposite binding at higher temperature. The RL parameter (0.375–0.193) lying between zero to unity validated the energetically favorable sorption, and contemplated a strong NB-nanocomposite interaction14 probably accounting for high percentage elimination of NB. The parameter, Qm is used to evaluate the sorption potential of a given adsorbent. A considerably higher adsorption efficacy (Qm) of 225.88 mg NB/g at 313 K relative to most described adsorbents in the literature for NB removal (Table 2) and various tree gum-based nanocomposites for other dyes confiscation (Table 3) validated the admirable sorption effectiveness of GG-PAAm/Er2O3 NC.
Isotherm plots: (a) Langmuir, (b) Freundlich, (c) Temkin, and (d) D-R for NB uptake.
Freundlich isotherm contemplates adsorption onto the surface of adsorbent with heterogeneous distribution of binding sites. It also illustrates the sorption as non-ideal and reversible phenomena where sorbent has non-uniform affinity leading to the multilayer sorption80. The magnitude of the parameter, nf is an indicator of the sorbent surface heterogeneity, and its value close to unity expresses a higher surface heterogeneity. The value of 1/nf is a measure of favorable, unfavorable or irreversible sorption process. The 1/nf value < 0.5 specifies the facile sorption, 1/nf > 1.0 denotes cooperative adsorption, while 1/nf > 2 depicts that NB is hardly sorbed54. The values 1/nf below 0.5 (0.445–0.475) together with relatively higher Kf (57.88–61.44 (mg/g)(L/mg)1/nf) (Table 1) supported the positive and favorable sorption of NB. The increasing trend in Kf with rise in temperature confirmed the endothermic trait of sorption.
Temkin isotherm model is employed to investigate the interaction between sorbate and sorbent81. It takes into account that heat of adsorption for pollutants diminishes linearly instead of logarithmically with an increase in coverage of the nanocomposite surface. The equilibrium binding constant, Kt (L/g) values (4.84–5.46) displayed an incremental change with rise in temperature (303–313 K) pointing out a relatively enhanced electrostatic interaction between nanocomposite surface-sites and NB molecules at high temperature. An increment in the bT values, which is related to the heat of sorption from 0.024 to 0.041 kJ/mol (Table 1) testified to slightly higher bonding probability of NB at elevated temperature (313 K). Further, the endothermic physisorption of NB was evidenced by the positive bT values below 8 kJ/mol, which is confirmed by the pertinent thermodynamic parameter (∆H°)82.
Dubinin–Radushkevich (D–R) isotherm model adopts a pore filling sorption mechanism with multilayer character involving van der Waals interaction and is usually used to recognize the mode of adsorption, that is, physical or chemical83. It also provides reasonable evidence about the adsorption mechanism with possible distribution of energy onto non-homogenous surface of adsorbent. The mean free energy of adsorption (E) was deduced using the equation, E = (frac{1}{(2upbeta )frac{1}{2}}) from the value of β (mol2/kJ2), estimated from Eq. (4). If the mean free energy, E is 1–8 kJ/mol then physical interaction governs the adsorption mechanism, while that between 8 and 16 kJ/mol indicates ion-exchange phenomenon. However, E > 16 kJ/mol specifies chemical interaction. The values of E equal to 0.081–0.091 kJ/mol advocated physisorption. The calculated values of qD (mg/g) were 139.74, 143.85, and 149.77 at 303, 308 and 313 K, which are in agreement with the similar trend in Qm values obtained using the Langmuir isotherm plot.
The estimation of excellent fit model is described on the basis of lower SEE and R2 values close to unity. It was concluded from Table 1 that Langmuir model with R2 near to one (0.975–0.995) and lower SEE (0.628–0.258) as compared to the values recorded for other isotherm models provided the best correlation of the data. The conformity of the equilibrium data to the Langmuir model implicated that the removal procedure occurred onto homogeneous surface of nanocomposite with the formation of monolayer coverage of NB molecules.
The adsorption kinetic models provide significant information on the rate of contaminant adsorption. These models are utilized to illustrate the experimental data to conclude the mechanism for the adsorption of contaminants from aquatic system at adsorbent–adsorbate interface. To interpret the adsorption procedure, the kinetic data was examined by pseudo-first order84 and pseudo-second order85 models by employing Eqs. (5) and (6), respectively:
where k1 (1/min) and k2 (g/mg/min) are adsorption rate constants for pseudo-first order and pseudo-second order kinetic model, respectively. The qe and qt signify adsorption capability of NB at equilibrium and time t, respectively. The values of k1, k2 and qe were determined for different initial NB concentration from the slope and intercept of the plot of qt versus t (Fig. 7a,b), and are presented in Table 4 along with R2 and SEE. The pseudo-second order model approved the best depiction of sorption data based on higher R2 (0.925–0.978) and lower SEE values (0.077–0.136), which indicated that the removal of NB by nanocomposite was affected by the number of active binding sites instead of the initial NB concentration. The appropriateness of the pseudo-second order kinetic model in describing the experimental data recommended that the rate limiting step for NB confiscation by GG-PAAm/Er2O3 NC probably involved chemisorption mechanism. The decrease in the magnitude of k2 values with increment in initial NB concentration (0.069–0.059 g/mg/min) (Table 4) signified rapid adsorption at lower concentration, which could be ascribed to the lesser competition faced by NB molecules for surface active-sites suggestive of physisorption54.
Kinetic plots: (a) pseudo-first order, (b) pseudo-second order, (c) intraparticle diffusion, and (d) liquid film diffusion for NB removal.
Interpretation of kinetic data is crucial to conclude the adsorption procedure governing the rate controlling steps. Usually, the liquid film diffusion implicating external mass transfer of NB molecules from bulk solution to GG-PAAm/Er2O3 NC surface, intraparticle diffusion and interior pore diffusion are encompassed in the adsorptive scavenging of dyes. The Boyd liquid film and Weber-Morris intraparticle diffusion models86,87 are respectively expressed as Eqs. (8) and (7)
where ki (mg/g/min0.5) and kD (1/min) are rate constants for intraparticle and liquid film diffusion, respectively and Ci is the intercept expressing the boundary layer thickness.
The straight-line curves of either qt versus t0.5 (Fig. 7c) or –ln(1–F) versus t (Fig. 7d) (F = qe/qt) at 40 mg/L and 50 mg/L initial NB concentration with Ci = 0 delineate that the dynamics of the confiscation process of NB is controlled either by intraparticle or liquid film diffusion as the rate-limiting step. The intraparticle diffusion graphs, however, deviated from the linearity with high boundary layer contribution to the rate-controlling step (Ci = 45.65 and 58.48) indicating that it did not solely control the adsorption rate. Similarly, the liquid film diffusion plots also were not linear and did not pass through the origin, which precluded the liquid film diffusion as the only rate governing step. The values of ki (mg/g min0.5) and kD (1/min) were 0.306–0.302 and 0.047–0.066, respectively at the studied concentrations. It could, therefore, be inferred that the adsorption process of NB was controlled by both the diffusion mechanisms. However, based on R2 and SEE values for intraparticle diffusion (0.985–0.987; 0.016–0.020) and liquid film diffusion (0.989–0.991; 0.003–0.007), it could be concluded that the liquid film diffusion has a predominant role.
Arrhenius equation (lnk2 = lnA − Ea/RT) was employed to determine the activation energy (Ea) for the adsorption of NB onto GG-PAAm/Er2O3 NC surface, where k2 is the pseudo-second order rate constant. The slope of lnk2 versus 1/T plot, displayed in Fig. 8a, gave the precise value of Ea,, and is listed in Table 5. Activation energy provides an insight about the nature of adsorption i.e., physical, or chemical. The low values of Ea (5–50 kJ/mol) are indicative of physisorption, whereas the Ea between 60 and 800 kJ/mol recommend chemisorption. Low values of Ea usually imply a process controlled by diffusion and greater values signify the involvement of chemical reactions. Therefore, the determined value (Ea = 15.33) advocated physisorption, which is consistent with the values reported in literature, which are 13.2 kJ/mol for adsorption of methyl violet onto perlite88 and 19.25 kJ/mol for maxilon blue 5G on sepiolite89.
(a) Arrhenius plot, (b) van’t Hoff plot, (c) reusability potential of GG-PAAm/Er2O3 nanocomposite.
The enhancement in NB removal on increasing the temperature from 298 to 313 K represented an endothermic adsorption process. The thermodynamic parameters such as changes in free energy (ΔG°), entropy (∆S°), and enthalpy (∆H°) are used to determine the feasibility, spontaneity, and nature of the sorption procedure. The parameters, ΔG°, ∆H°, and ∆S° were evaluated by employing Eqs. (9) and (10):
where kc = qe/Ce, R = universal gas constant (8.314 J/mol K), and T = absolute temperature (K).
The slope and intercept of the linear curves of logkc versus 1/T (Fig. 8b) provided the ΔH° and ΔS° values, respectively (Table 6). It has been reported that ∆H° values of 2–10 kJ/mol designates physisorption mechanism involving van der Waals interactions, between 2 and 40 kJ/mol denotes hydrogen bonding, while that above 60 kJ/mol deduces chemisorption90. The positive ∆H° (15.35–17.87 kJ/mol) directed that the sorption process was endothermic and involved physisorption. Moreover, positive ∆S° (0.079–0.087 kJ/mol/K) (Table 6) suggested an elevated randomness indicating an escalated degree of freedom at solid–liquid interface. Similarly, a decrease in the negative values of ΔG◦ (− 8.41 to − 9.67 kJ/mol) with increment in temperature advocated that the adsorption was spontaneous and more favorable at higher temperatures. It could, therefore, be concluded that the mechanism of NB sorption onto GG-PAAm/Er2O3 NC was mainly governed by physisorption at the studied temperatures (298–313 K).
The intention of regeneration is not only to recover the removal efficacy of spent adsorbent, but also to recycle and reuse the valuable adsorbent for several series of sorption–desorption without loss of efficiency and stability, which might be helpful in sustainable management of the waste adsorbents, and would cut the overall treatment cost. Since operating solution pH had significant impact on NB confiscation by GG-PAAm/Er2O3 NC, it was essential to control pH during desorption. For regeneration investigation, 2.0 g/L of GG-PAAm/Er2O3 NC was agitated with NB solution (50 mg/L) for 1 h, then 0.1 mol/L NaOH was utilized as an eluent. Figure 8c demonstrates the extent of NB removal by GG-PAAm/Er2O3 NC up to fifth sorption–desorption cycles. After fifth cycle, the R% diminished from 96 to 79, which might be explicated to the large number of reversible surface-sites of GG-PAAm/Er2O3 NC. The regeneration results pointed out that the GG-PAAm/Er2O3 NC was recyclable and extremely effective in eliminating NB signifying the excellent potential relevance of the adsorbent at industrial scale.
Viscometry method was used to appraise the biodegradability progress of GG-PAAm and GG-PAAm/Er2O3 NC. The advances in biodegradation was examined by assessing the intrinsic viscosity after every five days. From Fig. 9, it was apparent that both GG-PAAm and GG-PAAm/Er2O3 NC were susceptible to biodegradation. The solution depicted degradation in 5–50 days as the solution disclosed a significant loss in the viscosity, which suggested the biodegradable nature of the biopolymers. However, GG-PAAm/Er2O3 NC portrayed lesser degradability relative to GG-PAAm matrix, which might be due to an increased mechanical strength owing to erbium oxide doping.
Biodegradation curves of GG-PAAm and GG-PAAm/ Er2O3 nanocomposite.
The functional groups on the surface of GG-PAAm/Er2O3 NC, and the initial pH of the dye solution play significant roles in the adsorption of NB. The adsorption of pollutants onto different adsorbents generally occurs through various interactions such as electrostatic, hydrogen bonding, dipole–dipole, van der Waals forces and π–π. To explicate the mechanism accountable for the confiscation of NB by GG-PAAm/Er2O3 NC, pH and FTIR studies were utilized. The percentage removal of NB with pH suggested electrostatic interaction as one of the mechanisms responsible for NB uptake. The slight change in the position of IR spectral peak from 3028 to 3181 cm−1 and change in intensity of the vibrational bands at 1654 and 1081 cm−1 specified that the relevant functional groups were involved in the adsorption procedure through hydrogen bonding. However, the change in the peak intensities of O–H, C–O, and C–N groups confirmed the interaction of NB with existing functional groups. It might, therefore, be inferred that electrostatic interaction, hydrogen bonding and π–π interactions were mainly involved in the adsorption of NB onto GG-PAAm/Er2O3 NC surface. The plausible mechanism of NB sequestration by the GG-PAAm/Er2O3 NC is schematically illustrated in Scheme 1.
Schematic representation of the plausible interactions of NB with GG-PAAm/Er2O3 nanocomposite.
A novel GG-PAAm/Er2O3 nanocomposite with admirable adsorption capacity (225.88 mg NB/g) was efficaciously fabricated by an efficient, inexpensive, environmentally benign, and easy-to-use ultrasonic-assisted polymerization process. Fourier transform infrared spectroscopy, X-ray diffraction, scanning electron microscopy, energy dispersive X-ray, transmission electron spectroscopy, thermogravimetric analysis, specific surface area (SBET) and pHzpc measurements were used to characterize the synthesized nanocomposite, and was successfully employed for adsorptive elimination of nile blue from liquid phase. The surface area, pore volume and pore diameter of GG-PAAm/Er2O3 NC were 70 m2/g, 0.024 cm3/g and 5.796 nm, respectively. The adsorption parameters such as dose (0.8 g/L), concentration (80 mg/L), time (40 min) and pH (6) were optimized. The equilibrium data best fitted to the Langmuir isotherm model signifying homogenous sorption of NB onto the surface of GG-PAAm/Er2O3 NC. The high Qm value (225.88 mg NB/g) at 313 K validated better sorption competence of the GG-PAAm/Er2O3 NC for NB confiscation. The rate of NB sorption onto the sorbent surface was governed by pseudo-second order kinetic model with intraparticle and liquid film diffusion controlling the overall rate. The positive ΔH° (15.35–17.86 kJ/mol) suggested endothermic physisorption, whereas ΔS° (0.079–0.087 kJ/mol/K) indicated an increased randomness at the sorbent-NB solution interface. The negative ΔG° (− 8.41 to − 9.67 kJ/mol) governed the spontaneity and feasibility of the process. The regenerated adsorbent demonstrated good performance up to fifth cycles without much loss in efficiency, which implied that GG-PAAm/Er2O3 NC could be employed as an efficacious and potent adsorbent for cationic dyes including NB sequestration from waste water.
The datasets used and/or analysed during the current study available from corresponding author on reasonable request.
Lellis, B., Fávaro-Polonio, C. Z., Pamphile, J. A. & Polonio, J. C. Effects of textile dyes on health and the environment and bioremediation potential of living organisms. Biotechnol. Res. Innov. 3, 275–290 (2019).
Article Google Scholar
Dutta, S., Gupta, B., Srivastava, S. K. & Gupta, A. K. Recent advances on the removal of dyes from wastewater using various adsorbents: A critical review. Adv. Mater. 2, 4497–4531 (2021).
CAS Article Google Scholar
Khan, I. S., Ali, M. N., Hamid, R. & Ganie, S. A. Genotoxic effect of two commonly used food dyes metanil yellow and carmoisine using Allium cepa L. as indicator. Toxicol. Rep. 7, 370–375 (2020).
CAS PubMed PubMed Central Article Google Scholar
Hussain, D., Siddiqui, M. F., Shirazi, Z. & Khan, T. A. Evaluation of adsorptive and photocatalytic degradation properties of FeWO4/polypyrrole nanocomposite for rose bengal and alizarin red S from liquid phase: Modeling of adsorption isotherms and kinetics data. Environ. Prog. Sustain. Energy 20, e13822 (2022).
Google Scholar
Subhan, H., Alam, S., Shah, L. A., Ali, M. W. & Farooq, M. Sodium alginate grafted poly (N-vinyl formamide-co-acrylic acid)-bentonite clay hybrid hydrogel for sorptive removal of methylene green from wastewater. Colloids Surf. A Physicochem. Eng. Asp. 611, 125853 (2021).
CAS Article Google Scholar
Siddiqui, M. F., Khan, E. A. & Khan, T. A. Synthesis of MoO3/polypyrrole nanocomposite and its adsorptive properties toward cadmium(II) and nile blue from aqueous solution: Equilibrium isotherm and kinetics modeling. Environ. Prog. Sustain. Energy 38, e13249 (2019).
CAS Article Google Scholar
Khan, T. A., Rahman, R. & Khan, E. A. Adsorption of malachite green and methyl orange onto waste tyre activated carbon using batch and fixed-bed techniques: Isotherm and kinetics modelling. Model. Earth Syst. Environ. 3, 38 (2017).
Article Google Scholar
Kumar, A. et al. Bio-inspired and biomaterials-based hybrid photocatalysts for environmental detoxification: A review. Chem. Eng. J. 382, 122937 (2020).
CAS Article Google Scholar
Kausar, A. et al. Preparation and characterization of chitosan/clay composite for direct rose FRN dye removal from aqueous media: Comparison of linear and non-linear regression methods. J. Mater. Res. Technol. 8, 1161–1174 (2019).
CAS Article Google Scholar
Liu, Z., Khan, T. A., Islam, M. A. & Tabrez, U. A review on the treatment of dyes in printing and dyeing wastewater by plant biomass carbon. Bioresour. Technol. 354, 127168 (2022).
CAS PubMed Article Google Scholar
Abukhadra, M. R. & Mohamed, A. S. Adsorption removal of safranin dye contaminants from water using various types of natural zeolite. Silicon 11, 1635–1647 (2019).
CAS Article Google Scholar
Mao, B., Sidhureddy, B., Thiruppathi, A. R., Wood, P. C. & Chen, A. Efficient dye removal and separation based on graphene oxide nanomaterials. New J. Chem. 44, 4519–4528 (2020).
CAS Article Google Scholar
Momina, S. M. & Isamil, S. Regeneration performance of clay-based adsorbents for the removal of industrial dyes: A review. RSC Adv. 8, 24571–24587 (2018).
ADS CAS PubMed PubMed Central Article Google Scholar
Dassanayake, R., Acharya, S. & Abidi, N. Recent advances in biopolymer-based dye removal technologies. Molecules 26, 4697 (2021).
CAS PubMed PubMed Central Article Google Scholar
Etemadinia, T., Barikbin, B. & Allahresani, A. Removal of Congo red dye from aqueous solutions using ZnFe2O4/SiO2/Tragacanth gum magnetic nanocomposite as a novel adsorbent. Surf. Interfaces 14, 117–126 (2019).
CAS Article Google Scholar
Kulal, P., Krishnappa, P. B. & Badalamoole, V. Development of gum acacia based magnetic nanocomposite adsorbent for wastewater treatment. Polym. Bull. 20, 1–28 (2021).
Google Scholar
Hassanzadeh-Afruzi, F., Maleki, A. & Zare, E. N. Efficient remediation of chlorpyrifos pesticide from contaminated water by superparamagnetic adsorbent based on Arabic gum-grafted-polyamidoxime. Int. J. Biol. Macromol. 203, 445–456 (2022).
CAS PubMed Article Google Scholar
Thakur, S. et al. Synthesis and characterisation of zinc oxide modified biorenewable polysaccharides based sustainable hydrogel nanocomposite for Hg2+ ion removal: Towards a circular bioeconomy. Bioresour. Technol. 20, 126708 (2022).
Article CAS Google Scholar
Dong, K., Xu, K., Wei, N., Fang, Y. & Qin, Z. Three-dimensional porous sodium alginate/gellan gum environmentally friendly aerogel: Preparation, characterization, adsorption, and kinetics studies. Chem. Eng. Res. Des. 20, 25 (2022).
Google Scholar
Kulal, P. & Badalamoole, V. Evaluation of gum ghatti-g-poly (itaconic acid) magnetite nanocomposite as an adsorbent material for water purification. Int. J. Biol. Macromol. 193, 2232–2242 (2021).
CAS PubMed Article Google Scholar
Mittal, H., Morajkar, P. P., Al Alili, A. & Alhassan, S. M. In-situ synthesis of ZnO nanoparticles using gum arabic based hydrogels as a self-template for effective malachite green dye adsorption. J. Polym. Environ. 28, 1637–1653 (2020).
CAS Article Google Scholar
Mittal, H., Babu, R., Dabbawala, A. A., Stephen, S. & Alhassan, S. M. Zeolite-Y incorporated karaya gum hydrogel composites for highly effective removal of cationic dyes. Colloids Surf. A Physicochem. Eng. Asp. 586, 124161 (2020).
CAS Article Google Scholar
Banerjee, S., Debnath, A. & Singh, V. Gum ghatti-alginate hybrid bead derived titania spheres for deep removal of toxic dye remazol brilliant violet from aqueous solutions. Environ. Nanotechnol. Monit. Manage. 15, 100459 (2021).
Google Scholar
Mittal, H., Al Alili, A., Morajkar, P. P. & Alhassan, S. M. Graphene oxide crosslinked hydrogel nanocomposites of xanthan gum for the adsorption of crystal violet dye. J. Mol. Liq. 323, 115034 (2021).
CAS Article Google Scholar
Sharma, B. et al. Titania modified gum tragacanth based hydrogel nanocomposite for water remediation. J. Environ. Chem. Eng. 9, 104608 (2021).
CAS Article Google Scholar
Saya, L. et al. Guar gum based nanocomposites: Role in water purification through efficient removal of dyes and metal ions. Carbohydr. Polym. 261, 117851 (2021).
CAS PubMed Article Google Scholar
Hussain, D., Khan, S. A. & Khan, T. A. Fabrication and characterization of mesoporous guar gum/NiWO4 nanocomposite for efficient adsorption of phloxine B and crystal violet from aqueous solution and evaluation of its antioxidant activity. Colloids Interface Sci. Commun. 44, 100488 (2021).
CAS Article Google Scholar
Sharma, G. et al. Guar gum and its composites as potential materials for diverse applications: A review. Carbohydr. Polym. 199, 534–545 (2018).
CAS PubMed Article Google Scholar
Yan, L., Chang, P. R., Zheng, P. & Ma, X. Characterization of magnetic guar gum-grafted carbon nanotubes and the adsorption of the dyes. Carbohydr. Polym. 87, 1919–1924 (2012).
CAS Article Google Scholar
Khan, T. A., Nazir, M., Ali, I. & Kumar, A. Removal of chromium(VI) from aqueous solution using guar gum–nano zinc oxide biocomposite adsorbent. Arab. J. Chem. 10, S2388–S2398 (2017).
CAS Article Google Scholar
Sime, M., Das, A., Galt, G., Hope, G. & Young, C. Studies on the nature of salicyl hydroxamate adsorption at the surface of neodymium oxide. J. Dispers. Sci. Technol. 40, 1488–1498 (2019).
CAS Article Google Scholar
Ahmadi, S. et al. Acid dye removal from aqueous solution by using neodymium(III) oxide nanoadsorbents. J. Nanomater. 10, 556 (2020).
CAS Article Google Scholar
Zhang, Y., Li, K. & Liao, J. Facile synthesis of reduced-graphene-oxide/rare-earth-metal-oxide aerogels as a highly efficient adsorbent for Rhodamine-B. Appl. Surf. Sci. 504, 144377 (2020).
CAS Article Google Scholar
Marhalim, M. A. A. et al. Enhanced performance of lanthanum orthoferrite/chitosan nanocomposites for adsorptive photocatalytic removal of Reactive Black 5. Korean J Chem. Eng. 38, 1648–1659 (2021).
CAS Article Google Scholar
Mudhulkar, R., Damarla, K. & Pappula, V. N. Preparation and characterization of carrageenan-embedded lanthanum iron oxide nanocomposite for efficient removal of arsenite ions from water. Anal. Methods 14, 449–459 (2022).
CAS PubMed Article Google Scholar
Olivera, S. et al. Cerium dioxide and composites for the removal of toxic metal ions. Environ. Chem. Lett. 16, 1233–1246 (2018).
CAS Article Google Scholar
Singh, J. & Dhaliwal, A. S. Effective removal of methylene blue dye using silver nanoparticles containing grafted polymer of guar gum/acrylic acid as novel adsorbent. J Polym. Environ. 29, 71–88 (2021).
CAS Article Google Scholar
Shruthi, S. B., Bhat, C., Bhaskar, S. P., Preethi, G. & Sailaja, R. R. N. Microwave assisted synthesis of guar gum grafted acrylic acid/nanoclay superabsorbent composites and its use in crystal violet dye absorption. Green Sustain. Chem. 6, 11 (2016).
CAS Article Google Scholar
Hiremath, J. N. & Vishalakshi, B. Evaluation of a pH-responsive guar gum-based hydrogel as adsorbent for cationic dyes: Kinetic and modelling study. Polym. Bull. 72, 3063–3081 (2015).
CAS Article Google Scholar
Hamza, M. F. et al. Phosphorylation of guar gum/magnetite/chitosan nanocomposites for uranium(VI) sorption and antibacterial applications. Molecules 26, 1920 (2021).
CAS PubMed PubMed Central Article Google Scholar
Sharma, G. et al. Pectin-crosslinked-guar gum/SPION nanocomposite hydrogel for adsorption of m-cresol and o-chlorophenol. Sustain. Chem. Pharm. 6, 96–106 (2017).
Article Google Scholar
Gopi, S., Rajeswari, A., Sudharsan, G. & Pius, A. Highly crosslinked 3-D hydrogels based on graphene oxide for enhanced remediation of multi contaminant wastewater. J. Water Process. Eng. 31, 100850 (2019).
Article Google Scholar
Rezk, M. Y., Zeitoun, M., El-Shazly, A. N., Omar, M. M. & Allam, N. K. Robust photoactive nanoadsorbents with antibacterial activity for the removal of dyes. J. Hazard. Mater. 378, 120679 (2019).
CAS PubMed Article Google Scholar
Sahoo, J. K. et al. Guar gum-coated iron oxide nanocomposite as an efficient adsorbent for congo red dye. Desalin. Water Treat. 20, 342–354 (2017).
Article CAS Google Scholar
Santoso, S. P. et al. TiO2/guar gum hydrogel composite for adsorption and photodegradation of methylene blue. Int. J. Biol. Macromol. 193, 721–733 (2021).
CAS PubMed Article Google Scholar
Vanaamudan, A., Sadhu, M. & Pamidimukkala, P. Chitosan-Guar gum blend silver nanoparticle bio nanocomposite with potential for catalytic degradation of dyes and catalytic reduction of nitrophenol. J. Mol. Liq. 271, 202–208 (2018).
CAS Article Google Scholar
Sharma, G. et al. Gum acacia-cl-poly (acrylamide)@carbon nitride nanocomposite hydrogel for adsorption of ciprofloxacin and its sustained release in artificial ocular solution. Macromol. Mater. Eng. 305, 2000274 (2020).
CAS Article Google Scholar
Sharma, G. et al. Fabrication and characterization of novel Fe0@Guar gum-crosslinked-soya lecithin nanocomposite hydrogel for photocatalytic degradation of methyl violet dye. Sep. Purif. Technol. 211, 895–908 (2019).
CAS Article Google Scholar
Pal, S. et al. Modified guar gum/SiO2: Development and application of a novel hybrid nanocomposite as a flocculant for the treatment of wastewater. Environ. Sci. Water Res. Technol. 1, 84–95 (2015).
CAS Article Google Scholar
Khan, S. A., Siddiqui, M. F. & Khan, T. A. Ultrasonic-assisted synthesis of polyacrylamide/bentonite hydrogel nanocomposite for the sequestration of lead and cadmium from aqueous phase: Equilibrium, kinetics and thermodynamic studies. Ultrason. Sonochem. 60, 104761 (2020).
CAS PubMed Article Google Scholar
Azad, F. & Maqsood, A. Fabrication, structural characterization, dielectric and electrical parameters of the synthesized nano-crystalline erbium oxide. Electron. Mater. Lett. 10, 557–563 (2014).
ADS CAS Article Google Scholar
Boushehrian, M. M., Esmaeili, H. & Foroutan, R. Ultrasonic assisted synthesis of Kaolin/CuFe2O4 nanocomposite for removing cationic dyes from aqueous media. J. Environ. Chem. Eng. 8, 103869 (2020).
Article CAS Google Scholar
Rahimi-Nasrabadi, M. et al. Statistical optimization of experimental parameters for synthesis of two efficient photocatalyst: Erbium carbonate and erbium oxide nanoparticles. J. Mater. Sci. Mater. Electron. 28, 15224–15232 (2017).
CAS Article Google Scholar
Khan, T. A., Nouman, M., Dua, D., Khan, S. A. & Alharthi, S. S. Adsorptive scavenging of cationic dyes from aquatic phase by H3PO4 activated indian jujube (Ziziphus mauritiana) seeds based activated carbon: Isotherm, kinetics, and thermodynamic study. J. Saudi Chem. Soc. 20, 101417 (2022).
Article CAS Google Scholar
Arshadi, M., SalimiVahid, F., Salvacion, J. W. L. & Soleymanzadeh, M. Adsorption studies of methyl orange on an immobilized Mn-nanoparticle: Kinetic and thermodynamic. RSC Adv. 4, 16005–16017 (2014).
ADS CAS Article Google Scholar
Foroutan, R., Mohammadi, R. & Ramavandi, B. Elimination performance of methylene blue, methyl violet, and nile blue from aqueous media using AC/CoFe2O4 as a recyclable magnetic composite. Environ. Sci. Pollut. Res. 26, 19523–19539 (2019).
CAS Article Google Scholar
Kavitha, B. & Karthiga, R. Synthesis and characterization of CuWO4 as nano- adsorbent for removal of nile blue and its antimicrobial studies. J. Mater. Environ. Sci. 11, 57–68 (2020).
CAS Google Scholar
Esvandi, Z., Foroutan, R., Peighambardoust, S. J., Akbari, A. & Ramavandi, B. Uptake of anionic and cationic dyes from water using natural clay and clay/starch/MnFe2O4 magnetic nanocomposite. Surf. Interfaces 21, 100754 (2020).
CAS Article Google Scholar
Foroutan, R., Peighambardoust, S. J., Esvandi, Z., Khatooni, H. & Ramavandi, B. Evaluation of two cationic dyes removal from aqueous environments using CNT/MgO/CuFe2O4 magnetic composite powder: A comparative study. J. Environ. Chem. Eng. 9, 104752 (2021).
CAS Article Google Scholar
Zubir, M. H. M. & Zaini, M. A. A. Twigs-derived activated carbons via H3PO4/ZnCl2 composite activation for methylene blue and congo red dyes removal. Sci. Rep. 10, 1–17 (2020).
Article CAS Google Scholar
Kushwaha, R., Garg, S., Bajpai, S. & Giri, A. S. Degradation of Nile blue sulphate dye onto iron oxide nanoparticles: Kinetic study, identification of reaction intermediates, and proposed mechanistic pathways. Asia-Pac. J. Chem. Eng. 13, e2200 (2018).
Article CAS Google Scholar
Işikver, Y. Removal of some cationic dyes from aqueous solution by acrylamide- or 2-hydroxyethyl methacrylate-based copolymeric hydrogels. Fibers Polym. 18, 2070–2078 (2017).
Article CAS Google Scholar
Nayak, A. K. & Pal, A. Statistical modeling and performance evaluation of biosorptive removal of Nile blue A by lignocellulosic agricultural waste under the application of high-strength dye concentrations. J. Environ. Chem. Eng. 8, 103677 (2020).
CAS Article Google Scholar
Langmuir, I. The adsorption of gases on plane surfaces of glass, mica and platinum. J. Am. Chem. Soc. 40, 1361–1403 (1918).
CAS Article Google Scholar
İyim, T. B., Acar, I. & Özgümüş, S. Removal of basic dyes from aqueous solutions with sulfonated phenol–formaldehyde resin. J. Appl. Polym. Sci. 109, 2774–2780 (2008).
Article CAS Google Scholar
Maderova, Z., Baldikova, E., Pospiskova, K., Safarik, I. & Safarikova, M. Removal of dyes by adsorption on magnetically modified activated sludge. Int. J. Environ. Sci. Technol. 13, 1653–1664 (2016).
CAS Article Google Scholar
Safarik, I. et al. Magnetically-modified natural biogenic iron oxides for organic xenobiotics removal. Int. J. Environ. Sci. Technol. 12, 673–682 (2015).
CAS Article Google Scholar
Alipour, N. & Namazi, H. Removing Paraquat and Nile blue from aqueous solution using double-oxidized graphene oxide coated by polydopamine nanocomposite. Int. J. Environ. Sci. Technol. 16, 3203–3210 (2019).
CAS Article Google Scholar
Safarik, I., Horska, K. & Safarikova, M. Magnetically modified spent grain for dye removal. J. Cereal Sci. 53, 78–80 (2011).
CAS Article Google Scholar
Ghoochian, M. Adsorption of Nile blue A from wastewater using magnetic multi-walled carbon nanotubes: Kinetics and equilibrium studies. Iran. J. Toxicol. 10, 7–12 (2016).
CAS Google Scholar
Pandey, S., Do, J. Y., Kim, J. & Kang, M. Fast and highly efficient removal of dye from aqueous solution using natural locust bean gum based hydrogels as adsorbent. Int. J. Biol. Macromol. 143, 60–75 (2020).
CAS PubMed Article Google Scholar
Preetha, B. K. & Vishalakshi, B. Microwave assisted synthesis of karaya gum based montmorillonite nanocomposite: Characterisation, swelling and dye adsorption studies. Int. J. Biol. Macromol. 154, 739–750 (2020).
CAS PubMed Article Google Scholar
Chaudhary, S., Sharma, J., Kaith, B. S., Sharma, A. K. & Goel, A. Gum xanthan-psyllium-cl-poly (acrylic acid-co-itaconic acid) based adsorbent for effective removal of cationic and anionic dyes: Adsorption isotherms, kinetics and thermodynamic studies. Ecotoxicol. Environ. Saf. 149, 150–158 (2018).
CAS PubMed Article Google Scholar
Arun Krishna, K. & Vishalakshi, B. Gellan gum-based novel composite hydrogel: Evaluation as adsorbent for cationic dyes. J. Appl. Polym. Sci. 134(47), 45527 (2017).
Article CAS Google Scholar
Fosso-Kankeu, E., Mittal, H., Mishra, S. B. & Mishra, A. K. Gum ghatti and acrylic acid based biodegradable hydrogels for the effective adsorption of cationic dyes. Ind. Eng. Chem. Res. 22, 171–178 (2015).
CAS Article Google Scholar
Ghorai, S., Sarkar, A. K., Panda, A. B. & Pal, S. Effective removal of Congo red dye from aqueous solution using modified xanthan gum/silica hybrid nanocomposite as adsorbent. Bioresour. Technol. 144, 485–491 (2013).
CAS PubMed Article Google Scholar
Kaith, B. S., Jindal, R. & Sharma, R. Synthesis of a gum rosin alcohol-poly (acrylamide) based adsorbent and its application in removal of malachite green dye from waste water. RSC Adv. 5(54), 43092–43104 (2015).
ADS CAS Article Google Scholar
Sharma, G. et al. Fabrication and characterization of Gum arabic-cl-poly (acrylamide) nanohydrogel for effective adsorption of crystal violet dye. Carbohydr. Polym. 202, 444–453 (2018).
CAS PubMed Article Google Scholar
Bouaziz, F. et al. Efficiency of almond gum as a low-cost adsorbent for methylene blue dye removal from aqueous solutions. Ind. Crops Prod. 74, 903–911 (2015).
CAS Article Google Scholar
Freundlich, H. Uber die adsorption in losungen, zeitschrift fur phtsikalische chemie. Z. Phys. Chem. 62, 121–125 (1906).
Google Scholar
Temkin, M. I. Kinetics of ammonia synthesis on promoted iron catalysts. Acta Physiochim. URSS 12, 327–356 (1940).
CAS Google Scholar
Siddiqui, M. F. et al. A sugarcane bagasse carbon-based composite material to decolor and reduce bacterial loads in waste water from textile industry. Ind. Crops Prod. 176, 114301 (2022).
CAS Article Google Scholar
Dubinin, M. M. The equation of the characteristic curve of activated charcoal. In Dokl. Akad. Nauk. SSSR. 55, 327–329 (1947).
CAS Google Scholar
Corbett, J. F. Pseudo first-order kinetics. J. Chem. Educ. 49, 663 (1972).
CAS Article Google Scholar
Ho, Y. S. & McKay, G. Two-stage batch sorption and optimized design for dye removal. Chem. Eng. Trans. 76(Part B), 183–191 (1998).
CAS Google Scholar
Boyd, G. E., Adamson, A. W. & Myers, L. S. The exchange adsorption of ions from aqueous solutions by organic zeolites. II. Kinetics. J. Am. Chem. Soc. 69, 2836–2848 (1947).
CAS PubMed Article Google Scholar
Weber, W. J. Jr. & Morris, J. C. Kinetics of adsorption on carbon from solution. J. Sanit. Eng. Div. 89, 31–59 (1963).
Article Google Scholar
Doğan, M. & Alkan, M. Adsorption kinetics of methyl violet onto perlite. Chemosphere 50, 517–528 (2003).
ADS PubMed Article Google Scholar
Alkan, M., Doğan, M., Turhan, Y., Demirbaş, Ö. & Turan, P. Adsorption kinetics and mechanism of maxilon blue 5G dye on sepiolite from aqueous solutions. Chem. Eng. J. 139, 213–223 (2008).
CAS Article Google Scholar
Abbasi, N., Khan, S. A., Khan, T. A. & Alharthi, S. S. Statistical evaluation of liquid phase sequestration of acridine orange and Cr6+ by novel mesoporous glutamic acid-g-polyacrylamide/plaster of paris/riboflavin hydrogel nanocomposite. Environ. Res. 213, 113712 (2022).
CAS PubMed Article Google Scholar
Download references
The author (DH) thanks the Ministry of Tribal Affairs, Govt. of India for award of Senior Research Fellowship, while (SSA) acknowledges the financial support through Taif University Researchers Supporting Project number (TURSP-2020/90), Taif University, Taif, Saudi Arabia.
Department of Chemistry, Jamia Millia Islamia, Jamia Nagar, New Delhi, 110 025, India
Daud Hussain, Suhail Ayoub Khan & Tabrez Alam Khan
Department of Chemistry, College of Science, Taif University, P.O. Box 110999, Taif, 21944, Saudi Arabia
Salman S. Alharthi
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
D.H.: conceptualization, investigation, methodology, data curation, formal analysis, writing—original draft, S.A.K.: investigation, methodology, data curation, software, formal analysis, editing; T.A.K.: conceptualization, validation, editing, resources, supervision; S.S.A.: validation, editing, resources.
Correspondence to Tabrez Alam Khan.
The authors declare no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and Permissions
Hussain, D., Khan, S.A., Khan, T.A. et al. Efficient liquid phase confiscation of nile blue using a novel hybrid nanocomposite synthesized from guar gum-polyacrylamide and erbium oxide. Sci Rep 12, 14656 (2022). https://doi.org/10.1038/s41598-022-18591-0
Download citation
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-022-18591-0
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
Advertisement
Advanced search
© 2022 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.
India vs Pakistan, Asia Cup 2022 Highlights: Hardik Pandya finishes in style as India defeat Pakistan by 5 wickets – The Indian Express
India vs Pakistan, Highlights: Hardik Pandya’s all-round brilliance powered India to a thrilling five-wicket win over arch-rivals Pakistan in an Asia Cup game on Sunday. Pandya ended with figures of 3/25 in 4 overs while restricting Pakistan to 147 in 19.5 overs and then smashed 33 off 17 balls with the winning six in the last over bowled by left-arm spinner Mohammed Nawaz.
Ravindra Jadeja contributed with 35 off 29 balls. When India bowled, senior seamer Bhuvneshwar Kumar had the best figures of 4 for 26. When India chased, former skipper Virat Kohli scored 35 off 34 balls. Naseem Shah was Pakistan’s best bowler with figures of 2/27.
Brief Scores:
Pakistan: 147 all out in 19.5 overs (Mohammed Rizwan 43, Hardik Pandya 3/25, Bhuvneshwar Kumar 4/26).
India 148 in 19.4 overs (Hardik Pandya 33 no off 17 balls, Ravindra Jadeja 35 off 29 balls).
Follow IND vs PAK highlights below.
“In bowling my plans are simple. It’s important to assess conditions, situations and use the weapons – hard lengths are my strength. I use it wisely, put doubt in the batsmen, ask them to play wrong shots./
In batting over the years I have understood that if I stay calmer, I can execute my plans. Those 50-50 chances I take, if I stay calm, I can do it my way.
I always knew that there was one guy who was making his debut and there was one left-arm spinner. I fancied my chances. Even if there was 15 needed in last over, I would have fancied my chances.
I try to keep my mind clear. I know that bowler is under more pressure bowling the 20th over to me. It’s not ego thing. Just that spinner and all.
We knew that belief was always there in our group. No matter what the situation is we can pull it off. These things can happen often in the game. We can pull off games like that. We believe in quality of players. We want to give enough clarity to what they can do in middle.
It was challenging but I would take wins like this rather than one-sided wins. The bowlers are getting better over last few months. It’s about giving them confidence to deliver what they are expected to do. Yes they were challenged a bit but overcoming them will give confidence.
Hardik Pandya since he has made his comeback, he has been brilliant. He had a great IPL. When he was out of team, he got into fitness regimen, figured what he had to do with his body, hits 140 kmph now and has been brilliant for us.
He has been a lot calmer, quite confident youngster, knows what he want to do. He knows what sort of fields he wants. Very confident. Can bowl quick – the bouncers he bowled today.
It’s about understanding the game -and he does that now. He was calm in taking the team through. High-pressure game, 10 runs per over, and there are times when one could panic but he didn’t.
“We were a few runs short but made a comeback in our bowling. Our fast bowlers were good. Dhani stepped up (with batting) and he also bowled well. Unfortunately we couldn’t finish off.
On the last over bowled by the left-arm spinner: We were trying to take it deep. He bowled well and Hardik finished it well. Naseem was very good, the way he bowled and showed his aggression.
Thank you very much. We wanted to play till the end, they have a very good bowling attack, their fast bowlers don’t give anything away. I could have finished the game – left-arm spinner against left-hander, but Hardik played superbly. He (Pandya) came out and said he was going to play his shots, and happy that he stayed till the end.
Hardik Pandya knows he is a cool cat. At times, he lets people know it too. He hit the match-winning six, accepted a bow from Dinesh Karthik, looked up at the dressing room and signalled a ‘telephone ring’ kind of wiggle. The same wiggle that Shahnawaz Dahani does after taking wickets though perhaps there was no connection between the two here. As it would happen, Dhani was one of the first player he would bump into after that celebration, and there ensued a sweet moment between the two. Smiles and animated happy faces.
But the coolest moment happenned after the third ball of the final over. He had just moved outside leg stump and tried to smash it through the off side but couldn’t beat the infield. He then looked across at Karthik, shut his eyes and nodded his head – the Indian way to say ‘ all is well, hum hain na, it’s all cool, nothing to fret’. And next ball he smashed the orb over the wide long-on boundary. Ravindra Jadeja, who played a screamer of a knock, also raved about Pandya. “He was very clear when he came to bat. He said, I will play my shots no matter what.” (By Sriram Veera)
Mohammad Nawaz takes the wicket of Jadeja. He departs after a fruitful knock for his team. Dinesh Karthik comes in the middle. India are one hit away from victory. Hardik Pandya does it in style. IND 148/5 after 19.4 overs.
Hardik Pandya finishes things off in style 🎉#INDvPAK #AsiaCup2022 pic.twitter.com/O7e4UcWbL0
After an eventful 18th over which saw Pakistan penalized for slow over-rate, with an extra fielder inside the circle for the rest of the match, Hardik comes out all guns blazing in the 19th taking 14 runs from this one. What a thumping innings he has played. IND 141/4 after 19 overs.
Captains dread over-optimistic wicket-keepers. And Babar Azman has one in the form of Mohammad Rizwan. In Rizwan’s utopian world, anything that passes the batter is caught behind and a pad thud is equal to lbw. Plus he also has the acting skills that could get him a role in serious cinema. At least once an over he would break into a victory dance, it is such a convincing act with potential to con the most seasoned of umpires. So when Ravindra Jadeja missed a pull shot and the ball reached Rizwan’s hands, as expected the celebrations started behind the stumps. Babar looked confused. Having played with Rizwan for a long time, he knew he needed to be careful. There seemed a seed of doubt in the Pakistan captain’s mind but the prospect of getting the wicket of Jadeja was tempting. DRS showed Jadeja hadn’t edged. Rizwan was sheepish but he seemed to be indicating the din in the stadium. You can never win with Rizwan. (By Sandeep Dwivedi)
Naseem Shah picks the wicket of Suryakumar Yadav. Great bowling by the youngster!! India are four down now. Hardik Pandya comes to bat. Pandya gets it right on the bat and hits it towards the boundary line. IND: 97/4 after 15 overs
Bold of the day #INDvsPAK #INDvPAK #PakVsInd #PakistanFloods pic.twitter.com/H5pfVudYR8
Thus far, this partnership has been smart enough and even though the asking rate has touched 9, it isn’t a cause for much worry if these two can maintain their partnership. What is needed by Pakistan at this moment is a wicket and on cue, Babar brings back pace.
Amid the hullabaloo over Virat Kohli’s elongated lean patch, Rohit Sharma’s poor form has slipped under the radar a bit. The India captain had a laboured stay in the middle today. Rohit got stuck, had a few close shaves against Naseem Shah and Haris Rauf, attempted an ungainly heave against Shahnawaz Dahani and mistimed a whack off Mohammad Nawaz on the leg-side that landed between two fielders in the deep. He was on six off 15 balls, when the release shot came – a six over long-on against Nawaz. It proved to be a false dawn, as Rohit departed a couple of balls later, falling prey to a miscued lofted drive against the left-arm spinner. The opener made 12 off 18 balls. He has now scored just two half-centuries in his last 11 international innings. (By Shamik Chakrabarty)
In 2017, Pakistan’s left-arm spinner Mohammad Nawaz was suspended for two months after admitting to the charge of failing to report a suspect approach by a bookmaker in Pakistan Super League. “Nawaz was charged with a single violation, on one occasion, which he failed to disclose to the PCB vigilance and security department, full details of the approaches and invitations received by him to engage in corrupt conduct,” the PCB had said then.
He married next year to a Diagonistc Radiographer Izdihaar who lives in South Africa. Slowly, his cricketing life to returned to tracks as he made his way to Pakistan team in 2019. Perhaps, his biggest day in international cricket came today, in two balls to be precise, taking out Rohit Sharma and Virat Kohli.
He was just hit for a six over cow corner by Rohit but persisted in his teasing trajectory. First Rohit holed out to long-on and off the first ball of his next over, he had Virat Kohli, perhaps trying to go inside-out with spin to a ball from middle and leg, but instead holing out at long-off. (By Sriram Veera)
Jadeja walks down the pitch and hits it hard against Mohammad Nawaz as the ball goes towards the rope. The left-hander is playing well against the spinners. IND: 77/3 after 12 overs
Suryakumar Yadav sweeps it against Shadab Khan and gets a boundary. SKY times it well this time against the wrist spinner and gets the result. IND: 69/3 after 11 overs
Mohammad Nawaz takes the big fish now. Kohli looks to chip the ball over the covers but miscues it. Agony for the Indian team. Pakistan are back in the game now. Jadeja gets a six here. IND: 62/3 after 10 overs
Mohammad Nawaz comes to bowl now. Rohit Sharma slogs it this time and gets a six. Incredible batting by Rohit Sharma. 50 up for India. OMG!! Skipper Rohit Sharma departs. Huge blow!! IND: 50/2 after 8 overs
That’s a great shot by Virat Kohli against Haris Rauf. He jabs it well this time and gets the result. Another boundary (leg byes) for India towards the fine leg. IND: 38/1 after 6 overs
Kohli times it well this time against Dahani and gets a boundary. He executes it well towards the cover point and gets the result. IND: 29/1 after 5 overs
Virat Kohli pulls it this time and gets a six against Haris Rauf. He is looking in great touch now. Both the batters are batting well now. IND: 23/1 after 4 overs
Edged a wide one, dropped. Flirted with a well-outside off ball, missed. Inside-edge, missed the stumps. Mistimes a drive, falls just clear of mid-off. That was Virat Kohli’s initial few moments at Asia cup.
After a few months where he was left ruing his luck – once he even threw his head to the skies and cried out, “What else do you want me to do?”, is luck turning his way?
There was one stunning shot in all this iffy play and miss. The first bouncer from a Pakistani bowler, after India had picked so many with them, was eagerly awaited and when Shanawaz Dahani, called “trump card for Pakistan” by Waqar Younis, bounced one short, it didn’t quite rise as much as he perhaps envisaged.
Kohli swivelled into a fiery pull, way in front of the square, and with lot of time on his hands. Would Dahani bounce again at him, and this time a tad quicker and higher, at his head? (By Sriram Veera)
India vs Pakistan Live Score Streaming and Updates.
All eyes on India vs Pakistan match today: ‘Hype outside, normal inside’ – The first question to Pakistan captain Babar Azam at the press conference ahead of the India-Pakistan encounter on Sunday was seemingly harmless: “Dubai mein garmi zyada hain kya (Is it very hot in Dubai)?”
Babar promptly replied: “Garmi toh hain (It is hot)”. He then paused, seeming to realise the loaded undertone of the question, and said with a grin: “Hum zyada garmi nahin lete hain (We don’t feel the heat too much)”. He was referring to the inevitable “more-than-just-a-game” narrative drummed up whenever India and Pakistan duel, limited to multi-team tournaments these days. (More)
Semi-supervised classifier guided by discriminator | Scientific Reports – Nature.com
Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Scientific Reports volume 12, Article number: 14665 (2022)
Metrics details
Some machine learning applications do not allow for data augmentation or are applied to modalities where the augmentation is difficult to define. Our study aimed to develop a new method in semi-supervised learning (SSL) applicable to various modalities of data (images, sound, text), especially when augmentation is hard or impossible to define, i.e., medical images. Assuming that all samples, labeled and unlabeled, come from the same data distribution, we can say that labeled and unlabeled data sets used in the semi-supervised learning tasks are similar. Based on this observation, the data embeddings created by the classifier should also be similar for both sets. In our method, finding these embeddings is achieved based on two models—classifier and an auxiliary discriminator model, inspired by the Generative Adversarial Network (GAN) learning process. The classifier is trained to build embeddings for labeled and unlabeled datasets to cheat discriminator, which recognizes whether the embedding comes from a labeled or unlabeled dataset. The method was named the DGSSC from Discriminator Guided Semi-Supervised Classifier. The experimental research aimed evaluation of the proposed method on the classification task in combination with the teacher-student approach and comparison with other SSL methods. In most experiments, training the networks with the DGSSC method improves accuracy with the teacher-student approach. It does not deteriorate the accuracy of any experiment.
Classification is one of the common tasks that people try to solve using various automatic methods. These methods categorize samples into assumed categories based on the extracted features. Classification problems can refer to different data modalities, like images, video, text, or sound. The primary approach to pattern classification is to apply deep models because they offer automatic feature extraction; however, they need massive annotated datasets to train it. The quality of the model depends on how accurately data in the dataset is labeled. Manual annotations are precise but costly. Therefore other approaches are proposed1.
When a dataset is small, augmentation is frequently used. Data augmentation is a technique used to increase the amount of data by adding slightly modified copies of already existing data or newly created synthetic data from existing data. It acts as a regularizer and helps reduce overfitting when training a machine learning model. The use of the augmentation operations usually improves the quality of the model. It is relatively easy for natural images. In this case, augmentation includes geometric transformations, color space augmentations, kernel filters, mixing images, random erasing, feature space augmentation2. The disadvantage of augmentation is its need to adapt to the specificity of the input data. When we consider the natural language processing (NLP) domain, then more advanced and specific techniques must be applied3. Zhang et al.4 found that one of the valuable ways to do text augmentation is replacing words or phrases with their synonyms. For audio, we can apply noise injection, shifting time, changing pitch and speed, or nonlinear combination to construct new samples and linear method to construct labels5. For video, one considers augmentation on the spatial side, where the standard single-image augmentation strategies for image classification are used and on the temporal axis, when one randomly sub-samples a shorter video clip from the whole sequence6.
Another practical approach that is relevant when teaching models on a small amount of labeled data is transfer learning7. This approach uses a pre-trained model, i.e., a model trained on a larger, possibly similar, dataset. In this model, the last layers are replaced with new, randomly initialized layers. Then the parameters of new layers are trained on the available data while other parameters are frozen. Next, the whole network is tuned. Transfer learning gives better results when source tasks are related to target task8. In transfer learning, the body of the pre-trained network can be treated as a feature extractor or a function projecting the input data into a particular embedding space.
Data augmentation and transfer learning have a significant drawback when a task modality is restrictive, i.e., medical images. Data augmentation incorporates additional knowledge about the data that is not necessarily present in the data itself. When used on medical images, a common technique of horizontal image flipping might lead to training a predictive model that sees no anomaly in a patient with a heart located on the other side. Satellite image flipping might change the correlation between sunrise-sunset image features or features correlated with Coriolis Effect like hurricane spin. Similarly, the usage of transfer learning poses a risk of transferring any bias from the original domain to the final one. There are applications where the standard state-of-the-art techniques are sub-optimal or invalid to apply, and even a slight performance improvement is a significant step forward. We aim to develop a method for applications where such restrictions apply.
Recently, the somewhat old idea of pseudo labels included in semi-supervised learning (SSL)9 has been more and more popular in the case of deep learning10,11. SSL addresses training models with limited labeled patterns when a relatively large body of unlabeled patterns is available. The main goal of this task is to train a classifier model using both the unlabeled and labeled datasets. Ultimately, SSL allows for obtaining models with effects similar to the models trained in a fully supervised manner (Supervised Learning, SL), using a much smaller number of labeled samples based on the data representation extracted from unlabeled patterns.
Keeping in mind that
complex data augmentation needs a lot of domain knowledge,
it may be very difficult for some data modality12,
possessing a huge annotated dataset is very expensive,
we wanted to avoid these problems. It was the main motivation to perform our research. It aims to develop a new SSL method applicable to image, text, sound, and possibly other data types. Its essence is not based on data augmentation, as it is in many current methods13,14,15. In16 authors demonstrated that the key to obtaining good SSL performances is the quality of the sample perturbations.
While building the method, we assumed that the representation of labeled and unlabeled data comes from the same data distribution, i.e. embeddings generated by hidden layers of the classifier network, should be similar, regardless of whether the classifier input came from a labeled or unlabeled dataset. Inspired by the Generative Adversarial Network (GAN) learning process17, we used an additional discriminator model and adversarial training to make labeled and unlabeled embeddings more similar. Such an approach allowed us to avoid the difficulty of defining the similarity measure of such embeddings. The proposed method was named Discriminator Guided Semi-Supervised Classifier, abbreviated to DGSSC.
Summing up, the main contribution of this research is a new method of semi-supervised learning (SSL) that can be characterized as follows:
it profits from the assumption about the same data distribution of labeled and unlabeled datasets,
it is not based on data augmentation,
it can be applied to data of various modalities.
Most popular and best-performing SSL methods extensively use the augmentations and therefore incorporate external knowledge about the data in the training process. Our method stands out when compared to others as it applies to various challenging tasks. The method makes no assumptions about the data, which makes it completely independent of modality and the usage of augmentations.
The roots of our research lie in semisupervised learning, which is based on a small amount of labeled data with a large amount of unlabeled data used during training. Its primary goal is to avoid labeling a massive amount of data. The novelty of our method is the combination of adversarial training, first time introduced in GAN17 and SSL approach. Adversarial machine learning can be used in a variety of applications. This technique is most commonly used to execute an attack or cause a malfunction in a machine learning systems18,19. It is also applied in domain adaption, which refers to collecting the training and test sets from different sources or data shifts over time. In this case, there would be a discrepancy across domain distributions. At first glance, there is nothing in common with semisupervised learning and domain adaptation. However, considering unsupervised domain adaptation20 deeper, we can see that both problems are defined similarly. We have labeled data in the training set (source), and we want to predict labels for data in the testing dataset (domain). In this context, unsupervised adversarial domain adaptation methods will be interesting for comparison of similarity to our method. They minimize the distance between the target and source feature distributions (expressed by correlation distances or maximum mean discrepancy). The goal is to identify a feature space in which target and source domain samples are indistinguishable. We can mention here: CADA21, ACGAN22, DANN23, NoGRL24 methods. It is worth pointing out the differences of DGSSC in relation to them. The Consensus Adversarial Domain Adaptation method, abbreviated to CADA, is composed of four steps. In the second step, CADA uses an architecture that resembles the DGSSC architecture (two branches of the classifier for searching a domain-invariant feature space and discriminator). Still, both solutions differ in adversarial losses used to train the model. The Auxiliary GAN—ACGAN, unlike DCSSC, uses an adversarial image generation approach. DANN stems from Domain-Adversarial Neural Networks. The method does not use the idea of a discriminator. Instead, the authors introduce the gradient reversal layer (GRL) trick, ensuring that the feature distributions over the two domains are as indistinguishable as possible. In the NoGRL method, the gradient reversal layer is removed and substituted by the new confusion loss. The two last-mentioned methods differ in their nature compared to DGSSC.
The research described in this paper is contained in seven sections. In the next one, we give the background to the study by describing related works. The DGSSC method is shown in “Method”. Section “Experimental setup” describes experimental setup with the goal of experiments, datasets description, and experimental procedure. In section “Ablation study”, we analyze whether our assumptions align with the experiment. The central part of “Experiments” is focused on experimental results and discussion. The conclusions and plans for future research end the paper.
In the introduction, we discussed the relationship of the DCSSG with adversarial domain adaption methods. This section will show state of the art in the SSL context. It is worth starting from the taxonomy of SSL methods proposed in25. It distinguishes two main trends: one focuses exclusively on labeling data not yet recorded (transduction approaches) and the other, which aims to develop a model that works for each new input data (inductive approaches). Then inductive methods are further split into unsupervised preprocessing, wrapper methods, and intrinsically semi-supervised.
Unsupervised preprocessing separates the classification task from representation (manifold) training. In the first step, the methods perform preprocessing in three ways by: unsupervised feature extraction, clustering, or supporting the initialization of weights of the target model. Only in the second step, they train the model in a supervised way using an annotated set. The primary example of feature extraction is searching data representation using Autoencoder26. In the NLP domain, we can mention the Masked Language Modelling method that learns to predict masked token representation based on other tokens in the sentence. They are very efficient in the fine-tuning models in the text classification27,28.
Intrinsically semi-supervised methods use unlabeled datasets directly in a loss function. It is opposed to unsupervised preprocessing methods where both datasets are treated separately. One category of intrinsically semi-supervised methods are generative models. They are characterized by teaching models that generate new data similar to the training data based on an unlabeled set. Modeling this process is possible thanks to General Adversarial Networks (GAN)17. They consist of two models: generator and discriminator. Training GAN is expressed by Eq. (1)
where V means the optimized function described by Eq. (2)
G presents the function performed by generator while D refers to the function of discriminator. Various extensions are added to the discriminator in semi-supervised learning to consider labels into the function D29,30,31. In this new form, discriminator categorizes a given sample into K + 1 classes (K classes stem from the solved problem, and an additional class denotes the data coming from the generator). In such a form, the discriminator addresses the semi-supervised learning problem by using unlabeled data in the learning process.
Another worth mentioning example in this group is using Siamese networks. Training of Siamese nets needs the loss function based on the distance between the inputs embeddings. By default, in supervised learning (SL), this feature allows the SL method to approach the embeddings of samples from the same class while moving the embeddings of samples from other classes away by at least a certain distance (usually Euclidean or cosine) called margin. This technique of training models is called contrastive learning. The use of contrastive methods in SSL is based on a classifier that operates on the extracted data representation, as in the unsupervised pre-processing methods. Siamese network-based methods can be classified as manifold approximation methods25. Siamese-based methods often use substantial augmentation and a label propagation technique. An example is FixMatch13. In this case, to each unlabeled data classified above a certain threshold is assigned a pseudo-label. All labeled or pseudo-labeled data is heavily augmented and serves as data in contrastive learning.
By design, mixup methods are independent of augmentation. They imply fewer assumptions about the type of data they operate on, and in particular, do not require augmentation during their use in the SSL domain. The mixup technique is based on the linear interpolation of the training samples. An example of a method in the field of SSL is Interpolation Consistency Training (ICT)32.
Wrapping methods include training the teacher model T from labeled data, assigning pseudo labels to the data from the unlabeled set, and training the model student S on the obtained artificial labels. This process can be iterated assuming the previous model S as the new T model. It is teacher-student learning11. Examples that can be mentioned are the Noisy Student33 and self-distillation34 methods.
Readers interested in the taxonomy of deep semi-supervised learning methods based on loss function and model design are referred to35.
Graph based SSL methods rely on the geometry of the data represented by both labeled and unlabeled patterns. They create graph, where nodes correspond to training samples, edges represent similarities between them. It is possible to learn with very few labels how to propagate information through the graph by using the graph or manifold structure of data. These methods can be transductive as well inductive25. They consist of two steps: graph construction and label inference and can be further structured in each step. Details can be found in the survey paper36.
Our literature survey shows that SSL methods are mainly developed using one data modality, i.e., images (13,14,15,36,37,38), text (39,40,41,42,43) or sound (12,44). This unwritten standard runs the risk that developed approaches are effective only for this type of data. Many of the best-performing methods are based on the augmentation of data, which is modality-specific. This observation justifies the assumption made about building a generic method suitable for various modalities.
As we mentioned, the most effective SSL methods use augmentations or other transformations of the input data, introducing some knowledge about the data into the learning process. In our method, we want to make SSL techniques as much as possible independent of data augmentation to use it for various modalities (image, sound, text, etc.). Assuming that each training data from the labeled and unlabeled sets comes from the same data distribution, the embeddings (and predictions) generated by the model for labeled and unlabeled data types should also come from the identical distributions. This fundamental idea allows us to design our new method.
As the unlabeled and labeled samples are similar, i.e., come from the same distribution, all network activations on all layers should also be similar regardless if the sample is drawn from a labeled or unlabeled set. Also, we expect that if we used regular supervised learning with labeled data only, the network would likely overfit to samples seen during training. The overfitted model would make worse predictions for test samples than for the samples used for training, which means that the distribution of prediction vector values is different for training and test sets. Suppose the prediction vector values are of different distributions. In that case, we expect that we could notice distribution change for all network layers, on any depth, not only for the last prediction layer. We propose to pick a single layer of the classifier and use its activation distribution for the unlabeled points to detect when the overfitting occurs. More precisely, when the network is trained on a labeled set, we assume that if we monitor any layer activations for labeled and unlabeled sets, we could notice a growing distribution shift of the activations for labeled and unlabeled sets. We will call the activations of a chosen layer by embeddings. Further, we propose that if we encourage the embeddings for both sets to be more similar, we could somewhat prevent the overfitting. We notice that the training procedure of GAN networks is well suited for encouraging the embeddings of a labeled and unlabeled set to be similar.
The DGSSC method combines a discriminator D known from the GAN network and a standard classifier C. Its idea is presented in Fig. 1. In other SSL methods based on GANs, the generator generates samples while the discriminator evaluates them. Then, the generator model is discarded after training. In our method, the discriminator is trained to distinguish between the embeddings (hat{{textbf {e}}}_L) (for data stemming from the annotated set (D_{L})) and the embeddings (hat{{textbf {e}}}_U) of samples from the unlabeled set (D_U).
The classifier’s role is to classify samples from (D_{L}) dataset and to produce embeddings for samples from both datasets (D_{L}) and (D_U). The classifier is trained on the labeled set (D_L) in a regular, supervised way. It is also trained to generate embeddings for unlabeled data so that the discriminator would not distinguish from which dataset the embeddings are (labeled or unlabeled). Classifier creates embeddings that can be taken from any layer. We considered embeddings from the last layer before using the Softmax function in the experiments.
Data flow in the DGSSC method. (D_L) denotes the set of labeled samples ({textbf {x}}_L), (D_U) set of unlabeled samples ({textbf {x}}_U), C the classifier network and D the discriminator network. Classifier is trained in two modes. In supervised training it predicts output (hat{{textbf {y}}}). It also takes a part in adversarial training to produce embeddings (hat{{textbf {e}}}) of samples from both sets.
The classifier is trained to find better and better representation (embeddings) for the unlabeled samples to cheat the discriminator that the given sample is from the (D_L) dataset. The joint training of classifier and discriminator models encourages the improvement of classification performance.
Two loss functions are used during training the networks: (L_C) for the classifier and (L_D) for the discriminator. They are described in Eqs. (3) and (4), respectively.
The classifier loss (L_C) in Eq. (3) is composed of the (L_S) loss, which denotes a supervised learning part that updates C to classify labeled samples correctly, and the second component the (L_U) loss, which refers to the unsupervised part. It leads C to treat unlabeled samples as labeled ones by changing C to produce embeddings in a way that would make D classify unlabeled samples as labeled ones. They are defined in Eqs. (5) and (6).
In the above equations, H means cross-entropy, (hat{{textbf {y}}}_L) classifier predictions for labeled set, ({textbf {y}}_L) true labels, ({hat{{textbf {e}}}_U}) embedding of unlabeled sample, D represents the discriminator function. The discriminator recognises two classes—one corresponding to the labeled sample embedding when it outputs 1 and the second one for the unlabeled sample embedding when it outputs 0. In other words, 1 in Eq. (6) denotes target (label) used by discriminator for the labeled sample embedding.
The discriminator loss from Eq. (4) is composed of two components, as well. They are presented in Eqs. (7) and (8).
Variables here have the same meaning as in Eq. (6), i.e., H means cross-entropy, ({hat{{textbf {e}}}_U}) embedding of an unlabeled sample, ({hat{{textbf {e}}}_L}) embedding of a labeled sample. 0 and 1 denote targets (labels) used by the discriminator for the unlabeled and the labeled sample embedding, respectively. Moreover, D represents the discriminator function.
The (L_ {D_U}) and (L_ {D_L}) parts are used to train D to classify untagged and tagged samples, respectively. Even though in our case (L_U = -L_ {D_U}), we decided to leave these terms under separate symbols because it better reflects the intuition behind the method. In this case, it is easier to imagine another choice of loss functions for D and C , for instance, leaving the cross entropy for the training D in (L_ {D_U} ^ * = H (D ({hat{{textbf {e}}} _ U}), 0)) part and using mean square error for learning C in (L_U ^ * = MSE (D ({hat{{textbf {e}}} _ U}), 1)).
Equivalently, we can express the loss function in the minmax form (Eqs. 9 and 10)
where the adversarial setup arises from (H(D(hat{{textbf {e}}}_U), 0)) term relating to Eqs. (6) and (7).
The procedure for updating classifier and discriminator weights is presented in Algorithm 1. As an input, the algorithm takes classifier and discriminator models and two batches of samples—labeled and unlabeled. Also optimizers (opt_C, opt_D) for both models—classifier and discriminator are defined. As an output, the algorithm returns models of classifier and discriminator with updated weights.
This procedure is commonly referred to as a training step. It leads to updating models weights based on batches of samples drawn from labeled and unlabeled sets. When there are no more samples in one of the sets to pull the next batch, the set’s samples are reshuffled, and the next batch is drawn, independently of the second set’s state. The procedure is repeated until the monitored metric does not improve for the assumed number of consecutive measurements (in our experiments, it was equal to 8). This stop condition is commonly known as early stopping.
In Algorithm 1, lines 1 and 2 describe how the embeddings and predictions are calculated. Predictions stemming from unlabeled samples ((hat{{textbf {y}}}_U)) are not further used and are discarded. The loss calculation and weight update take place in lines 3-6 for the classifier, and in lines 7-10 for the discriminator.
Similarly to45, we enriched the above learning process by the teacher- student approach. It is a procedure that generally leads to improvement in many SSL techniques. We decided to adopt it as it does not incorporate any additional knowledge about the data into the training procedure and allows the method to stay independent from the data modality. At the same time, it does not constrain the usage of modality-specific procedures (for instance, augmentations). Its idea is illustrated in Fig. 2. The proposed procedure consists of four main steps, which make up one cycle O of the teacher-student method. In the first cycle, the classifier–teacher T and discriminator D models are initialized. Then in the first step, they are trained using the DGSSC method showed in the top of the Fig. 2. In the next step of the cycle, the trained teacher T is used to achieve pseudo-labels—embeddings of unlabeled set (D_U). It is represented by the rightmost element in Fig. 2. The new temporary dataset (D_U’) contains: sample ({textbf {x}}_U), and the corresponding pseudo-label, i.e., embedding (hat{{textbf {e}}}_U) assigned by the teacher network T. The dataset (D_U’) can be expressed as follows:
where P is the number of samples in dataset (D_U).
Next (bottom of Fig. 2), a new classifier-student S is initialized and trained to match its embeddings to teacher ones using the mean squared error loss function (L{_{MSE}}) (in Fig. 2 assigned as MSE):
where ({textbf {e}}_i) is the teacher embedding saved in the (D_U’) set and (hat{{textbf {e}}}_i) is the embedding of student classifier S. Lastly (the left element of Fig. 2) the trained student S is fine-tuned on the labeled set (D_L) in a supervised way using cross-entropy loss (L_S) defined in Eq. (5).
If it is the last cycle then the better classifier is returned, if not, student S becomes the new teacher model T and the cycle repeats without initializing a new teacher.
The diagram of model training in the teacher–student mode. The training comprises four main steps: starting from the one presented at the top—DGSSC procedure, followed by pseudo-labeling unlabeled set, knowledge transfer to a newly initialized model, and fine-tuning on labeled set. The objects on the blue arrows represent artifacts created by a given step used in the following step. The grey color denotes newly initialized models. T denotes the classifier–teacher, D discriminator, S the classifier–student, H cross-entropy loss, and MSE means squared error loss.
The detailed procedure of the teacher–student training scheme is presented in Algorithm 2. As an input, algorithm receives classifier (A_C) and discriminator (A_D) architectures, labeled (D_L) and unlabeled (D_U) datasets, the number of teacher–students cycles O, and the DGSSC method. As an output, it produces the trained classifier. In the first cycle, the classifier–teacher T and discriminator D are initialized (lines 3 and 5), and trained using the DGSSC method (line 6). The next step of the cycle is reflected in the line 7. Here, the trained teacher T is used to achieve pseudo-labels (hat{{textbf {e}}}_U) of the unlabeled set (D_U), and the pseudo-labels as the embeddings (hat{{textbf {e}}}_U) with corresponding input ({textbf {x}}_U) are saved into temporary dataset (D_U’) (line 7). The new classifier–student S is initialized (line 8), then trained to match its embeddings to teacher ones (line 9) using the mean squared error loss function (Eq. 12). The last step is described in line 10—the trained student S is fine-tuned on the labeled set (D_L) in a supervised manner using cross-entropy loss. In the last cycle, the better classifier is chosen and returned. In the case it is not the last cycle, the student S becomes the new teacher T (line 11), and the cycle repeats without initializing a new teacher (line 3 is skipped).
We declare that this paper describes our original research, and it is not considered for publication in other journals.
The goal of the experimental research is to check whether the proposed method of training a classifier jointly with the discriminator leads to improvement in classification accuracy in SSL setup. We will evaluate the DGSSC method in combination with the teacher–student approach allowing for gradual propagation of labels over the unlabeled set. As a reference, a teacher–student approach with a supervised classifier in place of the DGSSC method will be used. A supervised baseline is a classifier trained only on the labeled set without using teacher–student nor the DGSSC method. Supervised classifier architecture is the same as the architecture of the classifier in the DGSSC method.
All experiments were performed using PyTorch library46 and PyTorch Lightning47, a framework organising PyTorch code. The hyperparameters not mentioned in the paper, for instance, weight initialization schemes, were left as defaults from the PyTorch library, including the sourcing of datasets. The experiments were conducted on a single GPU station (Intel i9 CPU 64 GB RAM 1(times) Nvidia RTX 3090 GPU).
As we mentioned before, our goal was to verify whether the DGSSC method can be helpful for data in different modalities. Therefore in the experiments, we used six datasets of various modalities.
AG News, a text domain dataset based on AG’s corpus of news articles in a form described by4. The dataset consists of 127, 600 short news from four categories: world, sport, business, and science/technology.
IMDB48, a text domain dataset for binary sentiment classification. The dataset comprises 500, 00 movie reviews labeled as positive or negative.
Speech Commands, a sound domain dataset introduced in49 consists of 95, 848 recordings of 25 words that are not uniformly distributed in the dataset.
FINDSOUNDS, a sound domain dataset comprised of environmental sounds. The dataset was initially sourced from findsound.com search engine by the authors of50. It consists of 16, 930 sounds from 7 broad categories related to animals (e.g., cat, frog), people (e.g., coughing, laughing), nature (e.g., earthquake, ocean waves, flame), vehicle (e.g., car, helicopter, braking), noisemakers (e.g., alarm, bell, whistle), office (e.g., typing, printing) and musical instruments (e.g., bass, drum, synthesizer).
CIFAR-1051, an image domain dataset. It is a typical dataset in the evaluation of SSL methods. It consists of 60, 000 (32 times 32) colour images from ten classes, with 6000 images per class.
SVHN52, an image domain, digit classification dataset. It consists of 99, 289 (32 times 32) colour images of house numbers.
Detailed numbers of labeled, unlabeled, development and test dataset splits are presented in Table 1. All splits were stratified to preserve the relative frequency of class occurrences.
During the training process to detect model improvements, the classification accuracy metric on the development dataset was monitored. The accuracy achieved using the test dataset is used to report the final evaluation of the proposed model. Formally, accuracy is defined as shows Eq. ().
During the initial phase of method development, the following hyperparameters were proposed (they were not further optimised):
shared in all stages:
Learning rate decay: learning rate is divided by 10 after 5 consecutive validation measurements without improving the monitored metric,
Early stopping: if the monitored metric does not improve for 8 consecutive measurements, the training process is stopped.
Batch size: it is set to 100 based on the value proposed in32,
Validation measurement: it is performed on unlabeled dataset after every epoch in order to avoid ambiguity of epoch length of labeled and unlabeled sets,
optimizers:
Adam with (lr=0.001), (beta _1=0.9), (beta _2=0.999), default from the PyTorch library, for the transfer of knowledge while training models on pseudo-labels in teacher-student mode,
SGD with (lr=0.1), (momentum=0.9), (weight_decay=10^{-4}), as proposed in32, for the DGSSC classifier and baseline classifiers,
SGD with (lr=0.1), (momentum=0), (weight_decay=0), default from the PyTorch library, for the DGSSC discriminator.
In the method development phase, while working on the CIFAR-10 dataset, the three presented combinations of optimizers were tested for each of the training steps of the teacher-student model and baseline classifier, and the best ones were adopted for further experimentation.
Experiments on the text modality were performed using BERT architecture28. Its details are described in “Model used”. It is a relatively large and slow to train model, which forced us to reduce some of the hyperparameters as follows:
Batch size: it was set to 10 due to limitations of our GPU memory,
Learning rates: they have been divided by the factor of 100 for each optimizer to account for the smaller batch size, i.e.:
Adam with (lr=1text {e}-5) for the teacher–student mode,
SGD with (lr=0.001) for the DGSSC classifier and baseline classifiers,
SGD with (lr=0.001) for the DGSSC discriminator.
Validation measurement: it is performed after processing every 4000 sample from labeled and unlabeled sets, as awaiting for the processing of complete unlabeled set was infeasible because of computing resources.
In the case of text modality, remaining hyperparameters (for example early stopping or other optimizers’ hyperparameters) were left unchanged for this modality. In the development phase, we experimented with the division of all learning rates by the factor of 100 or 1000.
Experiments for all datasets in this section were repeated for different splits of the training dataset into labeled, unlabeled, and development datasets. There were three splits for the text modality and five for other modalities. Experiments on all datasets were made without using augmentation techniques.
In the case of the CIFAR-10 dataset, for comparison purposes with methods that use augmentation, we also did the augmentation in the form of zero padding by 2 pixels followed by random cropping back to the (32times 32) resolution and random horizontal flip with probability 0.5. The augmentations are adopted from32 and are used solely for better comparison with other methods.
We used 10 cycles in the teacher-student mode of training except for text modality where we train models for only 3 cycles as training large models based on BERT architecture was far slower compared to other experiments.
In DGSSC, two models play an essential role—discriminator D and classifier C. The discriminator model consists of five hidden layers of ten neurons with ReLU nonlinearity and one layer projecting to a scalar with Sigmoid nonlinearity. The architecture of ten hidden neuron layers was assumed to be both expressive enough and resulting in low model size as to make experiments fast. The discriminator input values were sorted as it was deemed to simplify the discrimination task as argued in the Ablation Study section.
The proposed method is independent of deep neural network architecture used as the classifier model. The choice of the classifier model depends on the problem domain/modality (image, text, sound).
As a text classifier we used the pretrained BERT model28 as proposed in42. It is a large model compared to models used in our experiments for other modalities. Detailed description of model layers is presented in Table 2. Training BERT model forced us to make changes in hyperparameters, primarily the batch size and corresponding learning rate to allow training the model on our station. We also reduced other hyperparameters influencing training time, such as the number of teacher–student cycles and frequency of measurements used for early stopping.
For the sound classification, the M5 network53 was used. It is a 1D-CNN model used commonly for the Speech Commands dataset in the PyTorch ecosystem. Detailed description of the model architecture is presented in Table 3.
For the image classification, the CNN13 network was used as proposed in32. Detailed description of the model is presented in Table 4.
Before presenting the experiments, to show intuitions standing behind the idea of our method in this section, we describe a simple example with the CNN13 network, which we call the baseline classifier. This network was trained in a supervised way using 4000 labeled samples from the CIFAR10 dataset (without augmentations). We analyzed the network activations on subsequent layers for 1000 from 4000 labeled samples used for training and another 1000 unlabeled ones (unseen by the classifier during training).
In Fig. 3 we present the network activations from subsequent layers projected to 2 dimensional space using T-SNE method (the parameters of T-SNE method were left as defaults from scikit-learn python package55: (perplexity=30.0, learning_rate=200), and random initialization), marked as labeled (orange) or unlabeled (blue).
T-SNE projection of activations of hidden layers for labeled and unlabeled samples for baseline classifier trained for CIFAR-10 dataset. From top-left shown are the predictions ((layer_{11})), logits ((layer_{10})), and other intermediate layers until the projection of activations of the first convolutional layer (bottom-right, (layer_0)). Labeled and unlabeled samples come from the same distribution therefore their activations should be indistinguishable, which is not satisfied in this case.
We can spot significant differences in how the network activations look like for labeled and unlabeled samples for layers from (layer_7) to (layer_{11}). When we assume that labeled and unlabeled samples come from the same distribution, their activations should be indistinguishable, which is not satisfied in this case.
The smaller the discriminator input, the faster the training is. Therefore we decided to use activations ({textbf {e}} in R^{10}) of (layer_{10}) as discriminator input. We will refer to them as sample embeddings. Inspired by the adversarial training from GAN17, we decided to use an auxiliary discriminator to encourage the embeddings for both labeled and unlabeled samples to be indistinguishable.
The model overfits to labeled samples used for training, resulting in the predictions for samples used for training having in general one value very close to 1 and the rest close to 0. The trained model makes less confident predictions for unlabeled samples, resulting in the highest value of the prediction vector being lesser in general than for the overfitted (labeled) samples, and probability mass will spread on other dimensions.
Let us notice that to recognize the sample as labeled one, the relationship between components of the prediction vector is essential. Its one value assigning the class should be almost equal to one, while others should be low. To distinguish the labeled samples from unlabeled, assigning to which class the sample belongs is not essential. Instead, it is crucial to identify this relationship between components of the prediction vector. In other words, rather than recognizing if an unlabeled sample belongs to one of 10 clusters formed by labeled sample predictions near points ((1,0dots 0), (0,1dots 0) dots (0,0dots 1)) in the prediction space, it is better to identify whether an unlabeled sample falls into a single cluster.
Figure 4a presents the scatter-plot of two most significant component values for labeled and unlabeled sample predictions (hat{{textbf {y}}}), which confirms our reasoning—many unlabeled sample predictions are easily separable from labeled overfitted ones. The last layer of all models considered in this paper is the Softmax operation, which preserves the ranking of its input values. We therefore suspect that (layer_{10}) embeddings ({textbf {e}}) may behave similarly to (layer_{11}) predictions (hat{{textbf {y}}}) under the sorting operation. As presented in Fig. 4b, we can notice that this is true as the labeled sample embeddings form a single cluster by observing the two most significant values of the embedding. This approach enables the discriminator model to operate on sorted embeddings and radically facilitates the task for the discriminator.
Two greatest components of activations for labeled (orange) and unlabeled (blue) samples for baseline classifier trained for CIFAR-10 dataset. Labeled samples (orange) are drawn on top of unlabeled ones (blue).
Training using our method led the classifiers activations to be more similar for labeled and unlabeled samples used in the training procedure. Fig. 5 presents T-SNE projection of classifier activations trained on the CIFAR-10 dataset (without the use of augmentations). Model activations for labeled and unlabeled samples from the last layers are noticeably more similar than in the case of the baseline classifier. More precisely, T-SNE projection of activations from (layer_7) forwards for the labeled and unlabeled samples are similar for the model trained with our method. Using the baseline classifier, the projection of unlabeled samples’ activations is more spread than labeled samples.
T-SNE projection of activations of hidden layers for labeled and unlabeled samples for classifier trained using the DGSSC method for CIFAR-10 dataset. From top-left shown are the predictions ((layer_{11})), logits ((layer_{10})), and other intermediate layers until the projection of activations of the first convolutional layer (bottom-right, (layer_0)).
To summarize, we can present the idea more formally, given the above intuition standing behind our approach. All problems considered in this paper refer to the multiclass classification; therefore, the labeled sample ({textbf {x}}_{textbf {i}} in {textbf {D}}_{textbf {L}}) has the output prediction of the Softmax layer – ((layer_{11})) in the form of vector ({{hat{{textbf {y}}}}_{textbf {i}}}=[{hat{{textbf {y}}}}_{i1},{hat{{textbf {y}}}}_{textbf {i2}},…,{hat{{textbf {y}}}}_{textbf {ic}}]), such that to following formulae is true:
where c is the number of classes. This relationship between the prediction vector components is valid for all labeled samples, independently of which k (which class) it is.
Having in mind that the set (D_L) is relatively small, with high probability the network is overfitted. Therefore considering embedding vector ({varvec{{hat{e}}_{textbf {i}}}}=[{varvec{{hat{e}}_{textbf {i1}}}},{varvec{{hat{e}}_{textbf {i2}}}},ldots ,{varvec{{hat{e}}_{textbf {ic}}}}]), for the same sample ({{textbf {x}}_{textbf {i}}}) from (layer_{10}), for embedding vector components the relationship shown in Eq. (14) can be expressed as follows:
This relationship, independently on k, characterizes the set of the labeled dataset and allows the discriminator to distinguish between labeled and unlabeled datasets.
To make the task for discriminator easier, we reorder the components of the embedding vector by sorting its components. For distinguishing labeled samples, it is not essential on which position is the highest value of the component.
Next, during adversarial training, the embeddings of unlabeled samples become more and more similar to the labeled ones, i.e., the relationship from Eq. (15) between components of the embedding vector arrives for unlabeled samples as well.
In order to validate the influence of the DGSSC method on classification accuracy in a semi-supervised setup, a supervised classifier in place of the DGSSC method was used. The results of the training procedure are presented in Fig. 6 and Table 5.
Experimental results of the proposed method and reference baselines. The lines represent the average from runs of each experiment and the lighter regions represent the standard deviation of the aggregated runs.
The subsequent rows of Table 5 show results of the DGSSC method without the teacher–student approach, the teacher–student approach with DGSSC representation learning at the first and last generation, the teacher–student approach with the supervised classifier replacing the DGSSC method, the supervised classifier baseline trained only on the labeled dataset and results of other methods. The results for the pure DGSSC method are equivalent to the performance of the first generation teacher when combining the method with the teacher–student approach. Subsequent columns in this table refer to various datasets.
The DGSSC method outperformed the reference baselines on datasets: Speech Commands, SVHN and CIFAR-10 with augmentations. No significant difference over the teacher–student-only approach was found on other datasets. In the case of non-image modalities, the usage of the teacher–student method did not improve the accuracy within consecutive cycles.
The analysis of the Speech Commands results shows that training with our method quickly reached its highest score (in the second cycle). In contrast, training without our method took approximately 4 cycles in the teacher–student mode to get the final score. We suppose that the characteristic of datasets plays an essential role in the training speed in terms of teacher–student cycles. The final score can be reached even in the first cycle, depending on the task.
Figure 6 visualizes the results for each cycle during training (horizontal axis). The vertical axis presents accuracy measured on the test dataset. The blue lines show experimental results of training DGSSC model combined with the teacher-student approach. The orange lines show the teacher-student approach with the supervised classifier in place of the DGSSC method. The green line represents the supervised baseline classifier. Dotted, dashed, and continuous lines represent the teacher, student trained on pseudo-labeled set, and the final student classifier results, respectively.
In the case of experiments on the CIFAR-10 without the use of augmentations, FindSounds, AG News, and IMDB datasets, the best results of our method overlap with the teacher-student baseline (teacher-student with supervised classifier in place of the DGSSC method), which is visible as overlapping error bands in Fig. 6. Out of those datasets, we see improvements only on CIFAR-10 without the use of augmentations and FindSounds datasets, where our method outperforms teacher-student baseline only in the first cycle of teacher-student training. The last four rows of Table serve as a general view of what is possible with the same combination of a dataset and labeled-unlabeled split. As noted in56, a direct comparison between the performance of different implementations shall not be made. With this in mind, we compare ourselves to MixText42 on text modality. We notice that our baselines and method (with nearly the same accuracy) achieved state-of-the-art results on AG News for 200 labeled samples per class. It is worth noting that experiments in the MixText paper were performed using only 5000 unlabeled samples during training which is significantly less than in our experiments. We decided to use all available unlabeled samples as it better refers to the SSL task (the SSL assumptions state that unlabeled samples are intrinsically available and the main cost of sourcing the data is the labeling process) and to be consistent with experiments on all other modalities. For the sound data classification, we found no adequate papers to compare. As a best effort, we can refer to50. Here, authors report Unweighted Average Accuracy (UAR) of (64.8pm 1.6%) on FindSounds dataset with the use of 4000 labeled examples. Our experiments achieved (UAR) (27.17pm 0.41%) for this dataset. For image modality, we compare ourselves to MixMatch14, ICT32, and FixMatch13 all of which outperform our results. Yet, given enough hyperparameter tuning and incorporating the same priors (i.e., in the form of adequate augmentation strategy), we suppose that our method would perform on par with those methods.
The results show that our method is promising in case of a challenging task, i.e., if it is hard to define augmentation or otherwise incorporate knowledge about the data. The proposed hyperparameters are reasonably robust to domain change, and it is unlikely that the usage of our method will degrade the final performance of a custom task. Further task specific fine-tuning would likely improve the performance.
Improvements of the proposed method are smaller than the best-performing state-of-the-art methods. However, those methods usually require extensive use of augmentations or, in other ways, incorporate knowledge about the data that is not necessarily present in the data itself. Our method is likely to improve results and is unlikely to introduce any bias into the model that would stem from assumptions required by other methods. We think the DGSSC method is beneficial for restrictive tasks requiring careful control over data assumptions.
In many experiments, training the networks with the DGSSC method improves accuracy when combined with the teacher–student approach of knowledge transfer. It does not deteriorate the accuracy in any of the experiments. The experiments involved two datasets per modality, single model architecture per modality, only one labeled-unlabeled split ratio per dataset, and datasets of different magnitudes. Therefore, it is impossible to conclude any significant insights on why the method performs on par with the teacher–student only method in some cases. As research requires intensive computing, we will leave this issue for clarification in the near future.
For the same reason, we conduct the experiments on the DGSSC method without tuning hyperparameters to the specific architecture and dataset combination. Further experiments with more computing resources could lead to more insight into the practicality of the method. The results suggest that even a single set of hyperparameters is enough to improve the accuracy for a wide range of applications.
The current version of our method leaves place for further improvements. Our nearest plans are listed below:
evaluation of the proposed method with the discriminator working on embeddings taken from different depths (layers) of the classifier networks,
swapping the discriminator with nonparametric loss functions such as Kullback–Leibler divergence,
usage of different cost functions in the knowledge transfer in the teacher–student approach (in this paper, only euclidean distance has been applied)
Finally, we would like to leave the reader with a thought that enforcing similarity of intermediate features between labeled and unlabeled samples in SSL is worth attention. We see the idea of leveraging invariance of model activation distribution by using adversarial training as a modality agnostic analog of the idea of leveraging data-specific invariance by using augmentations.
FindSounds dataset is available from the URL provided in50: http://zenodo.org/record/61295. All other datasets were sourced from the references provided by Pytorch47 library. SpeechCommands dataset: https://pytorch.org/audio/stable/datasets.html, text datasets: https://pytorch.org/text/stable/datasets.html, image datasets: https://pytorch.org/vision/stable/datasets.html.
Liu, B., Li, W.S. Lee, X.W.S. & Yu, P.S. Text classification by labeling words. in Proceedings of the Nineteenth National Conference on Artificial Intelligence, Sixteenth Conference on Innovative Applications of Artificial Intelligence (2004).
Shorten, C. & Khoshgoftaar, T. M. A survey on image data augmentation for deep learning. J. Big Data 6(60), 1115–2196 (2019).
Google Scholar
Lun, J., Zhu, J., Tang, Y. & Yang, M. Multiple data augmentation strategies for improving performance on automatic short answer scoring. Proc. AAAI Conf. Artif. Intell. 34(09), 13389–13396 (2020).
Google Scholar
Zhang, X., Zhao, J., & LeCun, Y. Character-level convolutional networks for text classification. in Advances in Neural Information Processing Systems, vol. 28. Curran Associates, Inc., Red Hook, NY, USA (Cortes, C., Lawrence, N., Lee, D., Sugiyama, M., Garnett, R. eds.) https://proceedings.neurips.cc/paper/2015/file/250cf8b51c773f3f8dc8b4be867a9a02-Paper.pdf (2015).
Wei, S., Zou, S., Liao, F. & Lang, W. A comparison on data augmentation methods based on deep learning for audio classification. Neural Netw. 1453, 012085. https://doi.org/10.1088/1742-6596/1453/1/012085 (2020).
Article Google Scholar
Yun, S., Oh, S.J., Heo, B., Han, D., & Kim J. Videomix: Rethinking data augmentation for video classification. arXiv:abs/2012.03457 (2020)
Pan, S. J. & Yang, Q. A survey on transfer learning. IEEE Trans. Knowl. Data Eng. 22(10), 1345–1359. https://doi.org/10.1109/TKDE.2009.191 (2010).
Article Google Scholar
Yang, Q., Zhang, Y., Dai, W., & Pan, S.J. Transfer Learning. 141–150. https://doi.org/10.1017/9781139061773.012 (Cambridge University Press, 2020).
Chapelle, O., Schölkopf, B., & Zien A. (eds.) Semi-Supervised Learning. http://dblp.uni-trier.de/db/books/collections/CSZ2006.html (The MIT Press, 2006).
Hinton, G.E., Vinyals, O., & Dean J. Distilling the knowledge in a neural network. arXiv:abs/1503.02531 (2015)
Pham, H., Dai, Z., Xie, Q., & Le Q.V. Meta pseudo labels. in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 11557–11568 (2021)
Grollmisch, S. & Cano, E. Improving semi-supervised learning for audio classification with fixmatch. Electronicshttps://doi.org/10.3390/electronics10151807 (2021).
Article Google Scholar
Sohn, K., Berthelot, D., Li, C.-L., Zhang, Z., Carlini, N., Cubuk, E.D., Kurakin, A., Zhang, H., & Raffel C. Fixmatch: Simplifying semi-supervised learning with consistency and confidence. arXiv: abs/2001.07685 (2020)
Berthelot, D. et al. MixMatch: A Holistic Approach to Semi-Supervised Learning (Curran Associates Inc., 2019).
Google Scholar
Berthelot, D., Carlini, N., Cubuk, E.D., Kurakin, A., Sohn, K., Zhang, H., & Raffel C. Remixmatch: Semi-supervised learning with distribution alignment and augmentation anchoring. arXiv: abs/1911.09785 (2020)
Ghosh, A., & Thiery A.H. On data-augmentation and consistency-based semi-supervised learning. arXiv :abs/2101.06967 (2021)
Goodfellow, I.J., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., & Bengio Y. Generative adversarial nets. in Proceedings of the 27th International Conference on Neural Information Processing Systems—Vol. 2. NIPS’14 . 2672–2680. (MIT Press, 2014).
Xu, H. et al. Adversarial attacks and defenses in images, graphs and text: A review. Int. J. Autom. Comput. 17, 151–178 (2020).
Article Google Scholar
Zhang, F., Chan, P. P. K., Biggio, B., Yeung, D. S. & Roli, F. Adversarial feature selection against evasion attacks. IEEE Trans. Cybern. 46(3), 766–777. https://doi.org/10.1109/tcyb.2015.2415032 (2016).
Article PubMed Google Scholar
Farahani, A., Voghoei, S., Rasheed, K., & Arabnia HR. A brief review of domain adaptation. arXiv:abs/2010.03978 (2020).
Zou, H. et al. Consensus adversarial domain adaptation. Proc. AAAI Conf. Artif. Intell. 33(01), 5997–6004. https://doi.org/10.1609/aaai.v33i01.33015997 (2019).
Article Google Scholar
Sankaranarayanan, S., Balaji, Y., Castillo, C.D., & Chellappa R. Generate to adapt: Aligning domains using generative adversarial networks. in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. 8503–8512 (2018)
Ganin, Y. et al. Domain-adversarial training of neural networks. J. Mach. Learn. Res. 17(59), 1–35 (2016).
MathSciNet MATH Google Scholar
Cherif, A., & Serieys H. Adversarial domain adaptation without gradient reversal layer. in ESANN (2020).
Van Engelen, J. E. & Hoos, H. H. A survey on semi-supervised learning. Mach. Learn. 109(2), 373–440 (2020).
MathSciNet Article Google Scholar
Hinton, G. E., Krizhevsky, A. & Wang, S. D. Transforming auto-encoders. In Artificial Neural Networks and Machine Learning—ICANN 2011 (eds Honkela, T. et al.) 44–51 (Springer, 2011).
Chapter Google Scholar
Radford, A. et al. Language models are unsupervised multitask learners. OpenAI Blog 1(8), 9 (2019).
Google Scholar
Devlin, J., Chang, M.-W., Lee, K., & Toutanova K. BERT: Pre-training of deep bidirectional transformers for language understanding. in Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Long and Short Papers). Vol. 1 . 4171–4186. (Association for Computational Linguistics, 2019).
Springenberg, J.T.: Unsupervised and semi-supervised learning with categorical generative adversarial networks. arXiv:abs/1511.06390 (2016)
Denton, E., Gross, S., & Fergus R. Semi-supervised learning with context-conditional generative adversarial networks. arXiv:abs/1611.06430 (2016)
Chen, X., Duan, Y., Houthooft, R., Schulman, J., Sutskever, I., & Abbeel P. Infogan: Interpretable representation learning by information maximizing generative adversarial nets. in Proceedings of the 30th International Conference on Neural Information Processing Systems. NIPS’16. 2180–2188 (2016)
Verma, V., Lamb, A., Kannala, J., Bengio, Y., & Lopez-Paz D. Interpolation consistency training for semi-supervised learning. in Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI-19. 3635–3641 (2019).
Xie, Q., Luong, M.-T., Hovy, E., & Le Q.V. Self-training with noisy student improves imagenet classification. in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2020).
Zhang, L., Bao, C. & Ma, K. Self-distillation: Towards efficient and compact neural networks. IEEE Trans. Pattern Anal. Mach. Intell.https://doi.org/10.1109/TPAMI.2021.3067100 (2021).
Article PubMed Google Scholar
Yang, X., Song, Z., King, I., & Xu Z. A survey on deep semi-supervised learning. arXiv: abs/2103.00550 (2021)
Lee, W.-Y., Hsieh, L.-C., Wu, G.-L. & Hsu, W. Graph-based semi-supervised learning with multi-modality propagation for large-scale image datasets. J. Vis. Comun. Image Represent. 24(3), 295–302. https://doi.org/10.1016/j.jvcir.2012.12.002 (2013).
Article Google Scholar
Rasmus, A., Berglund, M., Honkala, M., Valpola, H., & Raiko T. Semi-supervised learning with ladder networks. in Advances in Neural Information Processing Systems (Cortes, C., Lawrence, N., Lee, D., Sugiyama, M., Garnett, R. eds.). Vol. 28. https://proceedings.neurips.cc/paper/2015/file/378a063b8fdb1db941e34f4bde584c7d-Paper.pdf (Curran Associates, Inc., 2015).
Laine, S., & Aila T. Temporal ensembling for semi-supervised learning. arXiv:abs/1610.02242 (2017)
Miyato, T., Dai, A.M., & Goodfellow I.J. Adversarial training methods for semi-supervised text classification. in 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24–26, 2017, Conference Track Proceedings. https://openreview.net/forum?id=r1X3g2_xl (OpenReview.net, Online, 2017).
Sachan, D.S., Zaheer, M., & Salakhutdinov R. Revisiting LSTM networks for semi-supervised text classification via mixed objective function. in AAAI (2019).
Jo, H., & Cinarel C. Delta-training: Simple semi-supervised text classification using pretrained word embeddings. arXiv:abs/1901.07651 (2019)
Chen, J., Yang, Z., & Yang D. MixText: Linguistically-informed interpolation of hidden space for semi-supervised text classification. in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. 2147–2157. https://doi.org/10.18653/v1/2020.acl-main.194. https://aclanthology.org/2020.acl-main.194 (Association for Computational Linguistics, Online, 2020).
Baskaya, O. & Jurgens, D. Semi-supervised learning with induced word senses for state of the art word sense disambiguation. J. Artif. Intell. Res. 55, 1025–1058 (2016).
MathSciNet Article Google Scholar
Elibol, O.H., Keskin, G., & Thomas A. Semi-supervised and population based training for voice commands recognition. arXiv:abs/1905.04230 (2019)
Xie, Q., Hovy, E.H., Luong, M., & Le, Q.V. Self-training with noisy student improves imagenet classification. CoRR: abs/1911.04252. arXiv:1911.04252 (2019).
Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., Desmaison, A., Kopf, A., Yang, E., DeVito, Z., Raison, M., Tejani, A., Chilamkurthy, S., Steiner, B., Fang, L., Bai, J., & Chintala S. Pytorch: An imperative style, high-performance deep learning library. in Advances in Neural Information Processing Systems (Wallach, H., Larochelle, H., Beygelzimer, A., d’ Alché-Buc, F., Fox, E., Garnett, R. eds.). Vol. 32. 8024–8035. http://papers.neurips.cc/paper/9015-pytorch-an-imperative-style-high-performance-deep-learning-library.pdf (Curran Associates, Inc., 2019).
Falcon, W., et al. Pytorch lightning. in GitHub. Vol. 3. https://github.com/PyTorchLightning/pytorch-lightning (2019).
Maas, A.L., Daly, R.E., Pham, P.T., Huang, D., Ng, A.Y., & Potts, C. Learning word vectors for sentiment analysis. in Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies. 142–150. http://www.aclweb.org/anthology/P11-1015 (Association for Computational Linguistics, 2011).
Warden, P.: Speech commands: A dataset for limited-vocabulary speech recognition. arXiv:abs/1804.03209 (2018)
Han, W. et al. Semi-supervised active learning for sound classification in hybrid learning environments. PLOS ONE 11(9), 1–23. https://doi.org/10.1371/journal.pone.0162075 (2016).
CAS Article Google Scholar
Krizhevsky, A. Learning Multiple Layers of Features from Tiny Images. (University of Toronto, 2012).
Netzer, Y., Wang, T., Coates, A., Bissacco, A., Wu, B., & Ng A.Y. Reading digits in natural images with unsupervised feature learning. in NIPS Workshop on Deep Learning and Unsupervised Feature Learning 2011. http://ufldl.stanford.edu/housenumbers/nips2011_housenumbers.pdf (2011).
Dai, W., Dai, C., Qu, S., Li, J.B., & Das S. Very deep convolutional neural networks for raw waveforms. in 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). 421–425 (2017).
Salimans, T., & Kingma D.P. Weight normalization: A simple reparameterization to accelerate training of deep neural networks. in NIPS (2016).
Pedregosa, F. et al. Scikit-learn: Machine learning in Python. J. Mach. Learn. Res. 12, 2825–2830 (2011).
MathSciNet MATH Google Scholar
Oliver, A., Odena, A., Raffel, C., Cubuk, E.D., & Goodfellow I.J. Realistic evaluation of deep semi-supervised learning algorithms. in Proceedings of the 32nd International Conference on Neural Information Processing Systems. NIPS’18. 3239–3250 (Curran Associates Inc., 2018).
Download references
No funding was received to assist with the preparation of this manuscript.
Department of Arificial Intelligence, Wroclaw University of Science and Technology, wyb. Wyspiańskiego 27, 50-30, Wrocław, Poland
Sebastian Jamroziński & Urszula Markowska-Kaczmar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
Conceptualization: S.J. and U.M.K.; methodology: S.J. and U.M.K.; software: S.J.; validation: S.J. and U.M.K.; formal analysis: S.J. and U.M.K.; investigation: S.J.; resources: S.J. and U.M.K.; data curation: S.J.; writing-original draft preparation: S.J. and U.M.K.; visualization: S.J.; supervision: U.M.K.
Correspondence to Sebastian Jamroziński or Urszula Markowska-Kaczmar.
The authors declare no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and Permissions
Jamroziński, S., Markowska-Kaczmar, U. Semi-supervised classifier guided by discriminator. Sci Rep 12, 14665 (2022). https://doi.org/10.1038/s41598-022-18947-6
Download citation
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-022-18947-6
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
Advertisement
Advanced search
© 2022 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.
'La bohème' opens Kentucky Opera's 70th season in 2022 – Courier Journal
One of the world’s most famous and romantic operas will open the Kentucky Opera’s 2022-23 season this fall. Five performances of Giacomo Puccini’s “La bohème” will launch the 70th anniversary season of Kentucky Opera in grand fashion at The Brown Theatre, 315 W Broadway beginning Friday, Sept. 23.
First performed on February 1, 1896, in Turin, Italy, Classical Music magazine describes “La bohème” as “about as perfect as an opera can be. It’s concise, it’s packed with delicious melody, and it’s about being young and in love.”
“La bohème,” which was last performed in Louisville in 2013, has moved audiences for over a century with its luscious melodies and powerful storytelling of family and friendship among a group of struggling artists in Bohemian Paris.
“This opera has my heart,” Kelly Kitchens, stage director “La bohème.” “It was one of the first operas I ever saw as a young person, and I was transfixed. I still am. It is such a truly human story where we get a magnificent slice of these young people’s lives in which they wrestle with money, with love, with illness, with grief, and with joy.”
Kentucky Opera’s production will be performed over 120 minutes in four acts and will be sung in Italian with English captions. The opera will also feature members of Kentucky Opera’s Youth Opera Project and the River City Drum Corps.
You’ll have five opportunities to catch the Kentucky Opera’s production of “La bohème” including, Friday, Sept. 23 at 8 p.m., Sunday, Sept. 25 at 2 p.m., Tuesday, Sept. 27 at 1:30 p.m., Thursday, Sept. 29 at 8 p.m., and Sunday, Oct. 2 at 2 p.m.
Single tickets for “La bohème” start at $45.25, with discounts for students and groups, and are available now for purchase by visiting kyopera.org or by calling (502) 584-4500. Discount season subscriptions are also available to purchase starting at $110 and include “La bohème” as well as the 2022/23 season’s main stage performances of the tango opera “María de Buenos Aires” and Rossini’s “Cinderella.” A holiday production of “The Gift of the Magi” will be performed at the Opera Center, Kentucky Opera’s new home on Magazine Street.
All tickets are available now at kyopera.org. Student tickets are also available.
WHAT: The story is set in Paris around 1830 and shows the Bohemian lifestyle of a poor seamstress and her artist friends. Last performed by Kentucky Opera in September 2013.
CREDITS: Composer: Giacomo Puccini, Libretto by: Luigi Illica and Giuseppe Giacosa
WHEN: Sept. 23, 25, 27, 29 and Oct. 2
WHERE: Brown Theatre, 315 W. Broadway
WHAT: ‘María de Buenos Aires” tells the story of Maria, a “cursed” woman, who was seduced by the streets. In her quest for freedom, Maria turns her back on her former, innocent life and her world devolves into darkness as the gritty streets of Buenos Aires claim her as their own. However, Maria’s spirit survives, returning to haunt the city of her birth in death.Premiered: 1968, Buenos Aires, Argentina – Sala Paneta. This production will be a Kentucky Opera Premiere.
CREDITS: Composer: Ástor Piazzolla, Libretto by: Horacia Ferrer
WHEN: Nov. 11, 13
WHERE: Brown Theatre, 315 W. Broadway
WHAT: A chamber opera in one act. Based on the 1905 short story of the same title by O. Henry, the opera focuses on Jim and Della, a poor married couple who cannot afford to buy each other presents for Christmas. A Kentucky Opera premiere.
CREDITS: Music by David Conte, Libretto by Nicholas Giardini. Based on the 1905 short story by O. Henry
WHEN: Dec. 15-16, 18, 20, 22
WHERE: Kentucky Opera Center, 708 Magazine St.
WHAT: In this variation of the fairy tale, the wicked stepmother is replaced by a stepfather, Don Magnifico. The Fairy Godmother is replaced by Alidoro, a philosopher and tutor to the Prince. Cinderella is identified not by a glass slipper but by her bracelet. Premiered: 1817, Rome, Italy – Teatro Valle. Last performed by Kentucky Opera in 2005.
CREDITS: Composer: Gioachino Rossini, Libretto by: Jacopo Ferretti
WHEN: Feb. 24, 26, 2023
WHERE: Brown Theatre, 315 W. Broadway
Subscriptions for Kentucky Opera’s 2022-23 season are available now at kyopera.org or by calling 502-584-4500. Single tickets for individual performances will go on sale later in the year.
SLCC students preserve their Chicano culture through street art – Salt Lake Tribune
(Rick Egan | The Salt Lake Tribune) Beto Conejo paints a mural in collaboration with Deyvi Herrera on the side of a Salt Lake City building on Tuesday, Aug. 23, 2022.
Beto Conejo’s art features an amalgamation of cultures.
The bright colors and Indigenous symbols he paints on walls in Salt Lake City are an interpretation of his life. They include the sentiment of not truly belonging somewhere and wanting to enter unexpected spaces.
A self-taught muralist, Conejo paints what he knows — his Mexican roots, growing up in South Salt Lake and struggling with street violence. Through his art, he aims to offer a holistic vision of Chicano people like him.
“When people know me, they understand that that’s my background,” he said. “Once they see a piece of my art, whether it’s negative or positive, they get to see another piece of me.”
The art industry is a space he did not expect to access, Conejo said. But now, he is making a business plan to sell his artwork on his own terms.
Through Salt Lake Community College’s Everyday Entrepreneur program, Conejo is studying what it takes to found a media company and a studio to be able to elevate his culture. It’s not so much about making money, he said, but about doing significant work, and connecting with others.
During his coursework, he shaped his brand, Lords of the Night, with the hope of mixing different types of street art, such as murals, clothes, even skateboards.
Living in Utah, Conejo has been aware of the lack of access working people of color have to the outdoors. So he plans to expand his collection to include outdoor apparel for Latinx and Indigenous populations.
For him, this is a way to reclaim these populations’ narrative about their connection with nature, which has lost strength over time.
“Culture gets sold back to us cheap, and culture gets sold back to us through someone that’s not from it,” he said. “So we want to do it in a way that’s going to protect, preserve and explore both nature and culture at the same time.”
Coursework collaboration helps students shape their ideas and overcome barriers their businesses may face, said Jon Beutler, director of The Mill entrepreneurship center at SLCC’s Miller Campus in Sandy.
“It allows people from different backgrounds to work together and learn from each other and share ideas and thoughts,” Beutler said. “We had street artists from West Valley in the same room with a poet laureate with a Ph.D. in literature. And so, the conversations were more fun and vibrant.”
(Rick Egan | The Salt Lake Tribune) Beto Conejo paints a mural in collaboration with Deyvi Herrera on the side of a Salt Lake City building on Tuesday, Aug. 23, 2022.
The program gathers people from all over the world, and helps them conceptualize and launch different types of small businesses. One goal is to overcome the additional obstacles that people of color often face. In the case of some street artists, that could mean working harder to build trust.
“There’s a stigma with street art, people associated with tagging gangs, that sort of thing,” Beutler said. “But it’s not that; it’s just another form of art.”
And like any other art, it takes hard work to create, promote and sell.
(Trent Nelson | The Salt Lake Tribune) Daniel Morante, right, prepares to record a vocal track in the recording studio at Salt Lake Community College on Friday, Aug. 26, 2022. At left is Jesus Zavala.
Daniel Morante, a rapper and an audio engineering student at SLCC, knows all about laboring to improve his craft.
He took his initial hip-hop songwriting and rudimentary recordings to the next level when he was 17 and started riding the train from West Valley City to Salt Lake City to get to Spy Hop, a media learning center, where he learned audio fundamentals.
“I knew I was going to have to find a way to record the music that I was trying to create,” Morante said. “… Soon enough, I learned it was more complicated than I thought.”
Songs about socioeconomic issues and community concerns from Salt Lake County’s west side emerged from that effort.
“I don’t have to look outward when creating,” Morante said. “Everything I like to create is there in my area. I kind of like to create what represents me.”
(Trent Nelson | The Salt Lake Tribune) Daniel Morante records a track in the recording studio at Salt Lake Community College on Friday, Aug. 26, 2022.
He wants to continue recording music, but he also hopes to expand the network of artists in West Valley City.
While in the SLCC program, he formed West Valley Community, a brand meant to connect people from his area to resources to create art. His goal: Expose the working classes to fresh ways of thinking through audio engineering, music and community involvement.
“We provide everyday people for a sense of identity,” Morante said. “… I want to motivate my next generation that hasn’t developed or retained our culture.”
With his West Valley Community, Morante seeks to encourage young people to go to places that are available to them to expand their creativity.
“I hope that the brand becomes an icon for the youth and people in the community,” he said, “so when someone like me, a younger Chicano, looks at it, feels represented and feels like that it’s a positive message when they see it. Something that encourages them and that they want to be recognized with.”
Alixel Cabrera is a Report for America corps member and writes about the status of communities on the west side of the Salt Lake Valley for The Salt Lake Tribune. Your donation to match our RFA grant helps keep her writing stories like this one; please consider making a tax-deductible gift of any amount today by clicking here.
Donate to the newsroom now. The Salt Lake Tribune, Inc. is a 501(c)(3) public charity and contributions are tax deductible
Report a missed paper by emailing [email protected] or calling 801-237-2900
For e-edition questions or comments, contact customer support 801-237-2900 or email [email protected]
sltrib.com © 1996-2022 The Salt Lake Tribune. All rights reserved.
