This past summer, Euljeong Yun had just started an internship as a financial analyst with Buckhead Wealth Management when crisis struck. Another intern specializing in computer programming and development would not start as expected, leaving a gap in the team. Planned projects could not move forward without someone with coding proficiency. Although Yun didn’t consider herself an expert, she had gained confidence through the M.S. in Finance program at the Robinson College of Business and stepped into the vacant role.
“Even my boss acknowledged we had a tough problem to solve,” Yun said. “At first I wasn’t sure I could make it, but I focused on the task, and in a few weeks, I designed the company’s new analytics dashboard.”
Impressed by her skill range, Yun’s internship supervisor promoted her to quantitative financial analyst so she could take on more technical challenges.
“You don’t typically receive promotions or salary increases in an internship, but I did by demonstrating familiarity with financial analysis,” she said.
Learning and applying a diverse skill set is exactly why Yun chose Robinson in the first place. After graduating with a bachelor’s in economics from Yonsei University in South Korea, she researched many graduate programs.
“The M.S. in Finance program offered a perfect balance between technical chops, financial knowledge, and applied industry experience,” Yun said.
Yun particularly flourished in Rasha Ashraf’s Financial Data Analytics and AI and Machine Learning Applications in Finance courses, which provided hands-on experience with Python, SQL, SAS, and cutting-edge automation programs.
In keeping with her ambition to explore various facets of finance, Yun also has conducted scholarly research. She and Ashraf co-authored a paper, “ESG: A Tutorial using Bloomberg Data,” which will be presented at a conference next month. Also, Yun and a group of peers are using AI software to analyze Google data sets and assess how specific key search terms coincide with the rise and fall of cryptocurrency prices.
Yun currently works as a business analyst intern at Buildertrend, a construction project management software firm. As a member of the data monetization team, she is analyzing data sets with Python in order to increase company revenue.
“Robinson’s M.S. in Finance is especially strong for international students who may be hesitant to learn Python, SQL, or Tableau,” Yun said. “As a non-technical person, I was a little afraid to learn coding and other tools, but now I really enjoy it and can pursue a job that maximizes my financial knowledge and technical skills.”
Filed Under: Academic Unit News
More results…
Perimeter College Associate Dean Selected to Leadership Program
Jennifer Sherer Named Executive Officer for Entrepreneurship at Georgia State University
Lee to Discuss Exercise, Chronic Disease Reduction at 4th Annual Metzler Lecture
Perimeter College Students Draw First-Hand Career Insight from Visiting Chevron Engineers
Wells Fargo Financial Analyst Succeeded in Class and on the Volleyball Court
Send this to a friend
Master of Finance Student Impresses at Summer Internship – Georgia State University News
World Mourns Queen Elizabeth II, Britain's Bastion of Stability – The New York Times
World Mourns Queen Elizabeth II, Britain’s Bastion of Stability The New York Times
source
Second Annual Dome Fest West Announces Corporate Sponsors For This Year's October 6-9, 2022 Event – SHOOT Online
Designed to take full-dome content “to the next level,” the organizers of “Dome Fest West” have announced the Corporate Sponsors for their Second Annual/2022 event, which will begin on Thursday, October 6, 6-9pm/PT at the Orange Coast College Community Planetarium in Costa Mesa, CA – “the City of the Arts.” Dome Fest West 2022 will continue on Oct. 7 and Oct. 8, from 9am to 8pm, and on October 9, from 9am to 6pm. On Oct 9 at 7pm, a final award ceremony will take place. Tickets for Dome Fest West are available here: https://domefestwest2022.eventbrite.com
Dome Fest West (DFW) is an annual, 360-degree film festival and conference that brings together the filmmaking community, Hollywood filmmakers and up-and-coming immersive content creators from around the world. This year’s festival will screen numerous, fully-immersive, full-dome, 360-degree media shows, chosen as the best productions of 2021-22. The festival will also include new technology presentations, industry panels, interactive workshops, industry gatherings and parties. Event organizers expect up to 300 immersive filmmakers from around the globe to attend this year’s festival in-person as well as virtually.
The Corporate Sponsors for DFW’s 2022 event are:
Cosm: “Creators are the lifeblood of our industry and full-dome filmmakers are the early pioneers of immersive storytelling. Cosm is proud to support Dome Fest West and this incredibly vibrant and creative community,” Neil Carty, VP, Head of Cosm Studios & Labs at Cosm.
Cosm is a global technology company that brings experiences to life in immersive environments. The company helps its partners create spaces and content that blur the lines of real and virtual across three primary markets:
Sports & Entertainment; Science & Education; Parks & Attractions.
Cosm is currently in production on an enormous, multi-million-dollar new Sphere Venue which is being constructed in Hollywood Park, near SoFi Stadium in Los Angeles: https://www.cosm.com/los-angeles-venue-announcement/
https://www.cosm.com/
Orange Coast College Planetarium: All students of Orange Coast College, the site of the Planetarium which will play host to this year’s DFW, can get into any DFW screening free by showing their student ID.
https://prod.orangecoastcollege.edu/community/planetarium/index.html
New Audio Technology: New Audio Technology was founded in 2012 by Grammy-nominated immersive audio producer Tom Ammermann. With more than two decades of experience in immersive audio production, Tom leads his team to develop audio and music production and reproduction tools for Next Generation Audio (NGA) applications. This means immersive and interactive audio for mobile enjoyment, theaters, installations, games, music recordings, broadcast, streaming, live performances and full-dome/planetarium environments. https://www.newaudiotechnology.com/
LightSailVR: A boutique creative studio pushing the boundaries of immersive storytelling, LightSailVR specializes in creating immersive content with compelling characters and stunning visuals for both brands and original programming.
https://lightsailvr.com/index.php
Bridge Arts Media: DFW Co-founder Kate McCallum’s company Bridge Arts Media is a 360-degree transmedia arts, entertainment and media production company that develops and produces innovative properties designed for transmedia and cross-venue platforms.
https://bridgeartsmedia.com/
Vortex Immersion: DFW Co-founder Ed Lantz’s company lives on the edge of LIVE immersive experience development, specializing in the creation of dome venues and 360º distribution.
https://vortexdomes.com/
Experience 360°: DFW Founder Ryan Moore’s company Experience 360°, a Virtual Experience Design Studio that currently focuses on producing fully spherical, ultra-high definition 360° media for current VR devices, is donating funds and time, as well as processing of all the video presentations.
https://www.wemake360.com/
Festoon Software: Festoon Software, which eliminated the need for a physical dome or planetarium so content creators can develop and test their shows remotely from anywhere in the world, has donated its software for us by DFW’s judges. In addition, the company is also giving away a software subscription to all award winners.
https://www.festoonsoftware.com/
Nest Immersion: Nest Immersion is a software company that specializes in 360° content creation, software development, and immersive audio-visual installations. Nest will be giving away licenses to all award winners.
https://nestimmersion.ca/
Mesmerica: Mesmerica 360 is a fully immersive music and art cinematic projection show for planetariums and giant screen theaters by visionary musician James Hood, currently selling out in theatres across the United States. The show features 360 projections and stunning 7.1 surround sound. Mesmerica will be screening at OCC in November 2022.
https://www.jameshood.com/mesmerica360
Interesting Side Note:
During July 8-22, 2022, Dome Fest West presented the World’s Very First-Ever Dome Film Festival in the Metaverse. The two-week event was hosted in Domelandia, a metaverse world created by Vortex Immersion Media’s community circle on the AltspaceVR platform. The festival screened 17 films in 3 categories and received over 8,000 visits from participants across the globe.
The festival’s Executive Director Ryan Moore said, “The metaverse allows us to reach a much larger audience than our place-based festival. We love showcasing the work of these pioneering artists who are shaping what promises to become next-gen entertainment.”
Adds Worldbuilder and producer Ed Lantz, CEO of Vortex, “Immersive dome theaters offer powerful experiences that transcend ordinary TV, cinema and mobile media. That’s why Madison Square Garden Entertainment is sinking $1.7billion into the largest dome in Vegas. However, many people do not have access to dome theaters, which are largely confined to planetariums and science centers which prefer educational programming. Entering a metaverse dome with a $300 Meta Quest headset is the closest you can come to the real thing.”
About DOME FEST WEST FOUNDERS:
Ryan Moore
Dome Fest West Executive Director/Founder Ryan Moore is also the founder of Experience 360°, a Burbank-based Virtual Experience Design Studio that has been creating branded and narrative immersive content since 2015. Moore’s diverse career within the film production arena includes experimental branded content (for clients Disney, Toyota and GoPro;) TV commercials (for clients Pepsi and Jack In The Box;) music festival coverage (for client Fuse Media;) and Reality TV projects (including the series The Bachelor and The Bachelorette.
Ed Lantz
DFW Co-Founder Ed Lantz is an immersive media entrepreneur, engineer, producer and experience designer. Lantz left aerospace engineering in 1990 to transform traditional planetariums into immersive visualization environments, designing over a dozen dome theaters worldwide including Egypt’s Library of Alexandria Planetarium, Domodigital at Papalote Museo del Niño in Mexico City, and the National Space Centre in Leicester, UK. He sees modern digital domes as “digital temples of transformation” for their ability to combine art and science to create compelling immersive journeys, performances and experiences with positive impact. His company Vortex Immersion Media develops immersive media venues and experiences for clients and audiences worldwide.
Kate McCallum
DFW Co-Founder Kate McCallum is an award-winning multi-media producer/writer, with 20 years background at NBCUniversal and Paramount Studios. She is also the founder of c3: Center for Conscious Creativity, a 501c3 whose mission is to “Create a better future through the power of arts, media and storytelling,” and Bridge Arts Media, a media development and production company. McCallum produces full-dome and immersive content, served six years on the Producers Guild of America New Media Council, is a member of the TV Academy, and recently co-authored Handbook of Research on the Global Impacts and Roles of Immersive Media.
About DOME FEST WEST
Launched in Fall 2021, Dome Fest West is an immersive Dome film festival and conference designed to introduce the global immersive Dome filmmaking community to Hollywood’s new, up-and-coming content creators.
Presenting “The World’s Best Dome Media that Exists Today,” Dome Fest West annually showcases full-dome film screenings, new technology presentations, industry panels, interactive workshops and exclusive parties. Immersive filmmakers from around the globe attend the festival each year – an event which has been designed to take immersive Dome content “to the next level.”
Check out Dome Fest West on social media:
Tickets are now on sale for the Second Annual “Dome Fest West” which will begin on Thursday, October 6, from 6-9pm/PT at the Orange Coast College Community Planetarium in Costa Mesa, CA – “the City of…
The First Annual “Dome Fest West” which will kick-off at the Orange Coast College Planetarium in Costa Mesa, CA – “the City of the Arts” – on Friday, October 8 and continue until the evening of Sunday,…
The First Annual “Dome Fest West” which will kick-off at the Orange Coast College Planetarium in Costa Mesa, CA – “the City of the Arts” – on Friday, October 8 and continue until the evening of Sunday,…
In what organizers are billing as a “Groundbreaking Film Festival Event,” the First Annual “Dome Fest West” will kick-off at the Orange Coast College Planetarium in Costa Mesa, CA – “the City of the…
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.
‘A new place to explore’ – Martha's Vineyard Times
Featherstone welcomes Garden Gate Child Development Center to its new home.
For more than two decades, Garden Gate Childhood Development Center and Featherstone Center for the Arts have been working together to encourage creativity, exploration, and thoughtful expression in Island kids.
Now, the two organizations have solidified their long-cherished partnership, as Garden Gate relocates to the Featherstone campus to continue with their mission of allowing children to lead the way in their educational experience early on in life.
Apart from the sprawling property of grazing fields filled with animals and walking trails snaking through vast expanses of conservation land, the interior spaces of Featherstone also provide kids with some essential learning tools. Educators at Garden Gate treat classrooms as studios for creativity, and the new learning spaces at Featherstone certainly qualify.
The Times got to experience the new classroom spaces firsthand during an open house held by Garden Gate last week, and spoke with a number of teachers who will be working there.
“I think we are very lucky to have Featherstone. It feels like this space is just meant to have children in it,” Garden Gate lead teacher Delia Gibson said of the Featherstone Farmhouse, which the child center has named Studio 2. Gibson added that the inside of Studio 2 is open and airy, with lots of opportunity for activities like painting, sculpture, crafts, and more. She also said the space feels warm and inviting, and dodges the sometimes institutional look of conventional classrooms. Studio 1, known by Featherstone as the Pebble building, will serve as a classroom for toddlers 1½ to 3½ years old. Studio 2 will house the older students in preschool and kindergarten, ages 3½ to 6.
Josie Black, a Garden Gate parent and member of the childhood center’s board of directors, said she is “absolutely ecstatic” about the next level of collaboration between the two organizations, and feels like it brings the art and creativity-based elements of Garden Gate to a place where opportunities for creative expression are literally everywhere. “Having this campus as a resource is so amazing for the kids — my son loves to be outside, he loves to explore, he would spend hours outside just going around and having new experiences,” Black said.
Particularly coming out of COVID, Black said, it’s important for kids to have trusted and experienced educators who will give them chances to grow not just academically, but emotionally and creatively, as well. She said one of the most impressive things about Garden Gate is the teachers’ enthusiasm and ability to form strong and supportive connections with each individual child. “The teachers have taken the time to know everything about my son, and to really understand him and celebrate him as a person,” Black said. “It is beyond words how much time they put into my son and his happiness and well-being.”
Leigh Ann Yuen, co-founder and director of Garden Gate, said the entire staff is pinching themselves about the partnership, and everyone is excited to start a new chapter in education. “It feels like we were just always meant to be here. It feels like we were always here,” Yuen said.
According to Yuen, both the learning studios will act as temporary spaces while Featherstone works on a campus-wide master plan that includes alterations or additions to existing buildings. There is no definitive plan for a new Garden Gate building on the Featherstone campus, although Yuen said that process is still being worked out. “This really represents the first phase of a multistep process that will result in a permanent, long-term education space for us,” Yuen said.
As kids become more accustomed to the new classrooms, they will slowly start to gain more and more ownership over the space, which Yuen said builds confidence and allows kids to more freely follow their own interests. For Garden Gate, every area can be used for creative art, creative construction, creative drama — pretty much anything that has to do with kids finding their strengths, weaknesses, and passions through inventiveness and individuality. “One of the things that differentiates us from other kinds of childhood programs is we don’t teach children what to learn, we teach them how to learn. We are giving them the tools they need to explore and form their own relationship with the world,” Yuen said.
For Yuen, one of the biggest draws of the Featherstone campus is the expansive conservation lands and enriching outdoor activities that are made possible by the surrounding wilderness. After the pandemic, when so many kids were cooped up inside taking classes on Zoom and doing homework assignments online, Yuen said it was paramount for Garden Gate to provide as many active play experiences as possible.
“We really wanted to start nurturing and healing all the trauma that kids have endured,” Yuen said. “There is a quote that says, ‘Kids don’t bounce off the walls if you take away the walls.’ We have seen specific behaviors where kids would be kind of rough-and-tumble in the classroom, then you take them outside and they still are playing, but they have enough space to play freely and not interrupt anyone else’s learning time.”
Co-director and lead teacher Dawn Warner said Garden Gate is all about meaningful learning — the kind of education that is important to kids. Often, in a more traditional setting, kids have fewer sources of inspiration and motivation. Similarly, if a child isn’t inspired by or interested in a certain topic, they most likely won’t take to it as well. “When we fill every lesson with meaning, it sticks with the kids a lot more,” Warner said. “And this space will be a new place to explore that is very carefully curated to be calming, but also inspiring and highly conducive to learning.”
Garden Gate has purposefully overstaffed their classrooms this year. Studio 2 (for older kids) has 14 students enrolled, and three teachers. This means kids will receive more individualized education and direct attention from teachers. “If one student wants to do some painting, someone else wants to pretend to bake a cake, and another kid wants to learn how to work with clay, we have plenty of adults to facilitate all those activities,” Warner explained.
For executive director of Featherstone Ann Smith, bringing Garden Gate operations to the campus is the culmination of many years of close collaboration, and a natural match for both organizations. “We have hosted them for various exhibits and kids’ programs in almost every building on the campus at some point,” Smith said.
Right before COVID, in the summer of 2019, Featherstone hosted the Garden Gate 20th anniversary event and its graduation ceremony. Once the pandemic hit, capacity restrictions for classrooms forced the child center to look for alternative spaces. Yuen and Warner reached out to Smith about bringing the Garden Gate 3- to 5-year-old program to Featherstone, and they’ve been on the campus ever since. “We feel like we have a winning combination in terms of lifelong learning, arts education, and creative expression,” Smith said.
According to Smith, Garden Gate has always managed to integrate art galleries and exhibits into their curriculum, such as the “Art of Flowers” exhibit, and the current William Blakesley exhibit. Smith said when you incorporate the arts into learning and growth, it provides young students with the tools to interact with the world in an analytic and creative way. “This relationship is mutually beneficial, and we can’t wait for that connection to continue on into the future, and see all the wonderful things that Garden Gate has to offer,” she said.
Garden Gate is located on the Featherstone campus, at 30 Featherstone Lane, Oak Bluffs. Head to gardengatemv.org for more information, or to enroll your child in an educational program.
The MV Times comment policy requires first and last name for all comments.
1500 Sound Academy – Instructors, Administrators, & Executives (SG, TH, CN, JP) – Music Business Worldwide
1500 Sound Academy | Instructors, Administrators & Executives (Singapore, Bangkok, Beijing & Tokyo)
1500 Sound Academy (www.1500soundacademy.com) founded by Grammy-winning producers James Fauntleroy, Larrance Dopson, and entrepreneur Twila True, has been educating aspiring music professionals in writing, producing, arrangement, mixing, recording engineering, and artist branding since 2018. After success in the U.S., we are expanding into Asia.
Our first academy in Asia opened in Taipei in March 2022 to much fanfare. Plans for the coming months include campuses in Singapore, Bangkok, Beijing, Tokyo, and Mumbai. We are actively recruiting music industry professionals, educators, and administrators to join our mission of realizing the full potential of Asian youth. We currently have the following positions available.
Please submit your CV in confidence to [email protected]. Interviews are being conducted in person in Singapore, Bangkok, Beijing, and Tokyo, or remotely if you’re based elsewhere.
Music Business Jobs is the most effective way for the global music industry to find the best candidates for their open positions.
In addition to being seen by over 300,000 monthly unique readers online, our listings are also emailed direct to more than 50,000 professional subscribers every week – across the US, UK, Europe, Asia-Pacific and beyond.
Click through below for details on how to submit a Music Business Jobs listing – and learn why our advertisers ‘strongly recommend’ us to their peers…
Audio leak exposes IK’s narrative of foreign conspiracy: minister – Business Recorder
Subscribing is the best way to get our best stories immediately.
ISLAMABAD: Responding to the PTI chairman’s alleged audio leak, Interior Minister Rana Sanaullah said on Wednesday that the leak had exposed Imran’s “entire narrative of foreign conspiracy”.
Addressing a press conference, the Pakistan Muslim League-Nawaz (PML-N) leader accused the former prime minister of dividing the nation with his “lies”.
Sanaullah criticized the PTI chief over his continuous “lies” that his government fell prey to a foreign conspiracy.
“Today, his entire scheme has been laid bare,” he said.
He said it was important to put each and every word of the video before the public to “expose the fraud being done by this conspirator”.
After playing the leaked audio before the media, Sanaullah stated that the conversation was proof that “they [the PTI] were trying to create a plot.” He quoted Imran Khan from the audio leak, allegedly instructing his principal secretary Azam Khan to “just play with it.”
He further stated that the two “changed the minutes of the meeting and replaced them with a fabricated story that feeds the PTI’s narrative”. The PTI leadership is destroying the country with their “lies” for the last six months. During his tenure, he destroyed the country’s economy, the political culture and isolated Pakistan at the international level, which is why India decided to attack Kashmir, while our country was in a vulnerable state, he said.
He has caused irreparable damage to Pakistan politically, morally and economically, Sana maintained.
Sanaullah also reminded the PTI supremo of sending the cypher to Chief Justice of Pakistan Umar Ata Bandial back in April, saying: “You went to Supreme Court calling for an inquiry on this. You can now go and submit the plea and we will accept, concede and agree to it ourselves.”
Copyright Business Recorder, 2022
Dominic Fike Adds Second Los Angeles Show at the Wiltern – Grimy Goods
Because one Dominic Fike show is not enough! Dominic Fike has added a second Los Angeles show date, this time at the Wiltern for Saturday, Dec. 17. Tickets go on sale tomorrow, Friday Sept. 30 at 10AM. There is no presale for this one, so be sure to set you alarm and login to your Ticketmaster account at least 30 minutes before tickets go on sale and hop in the digital queue.
GET TICKETS TO DOMINIC FIKE AT THE WILTERN
Dominic Fike‘s Nov. 12 concert date at the Hollywood Palladium sold-out quickly! Don’t miss out on this second chance to catch him live!
As Part of his ‘Out of Order’ North American tour, Dominic Fike will headline the Hollywood Palladium on Saturday, Nov. 12. and Saturday, Dec. 17 In addition to Dominic Fike’s Los Angeles concert date, the Euphoria heartthrob has marked 28 concert dates including shows at The Masonic in San Francisco, the Roseland Theater in Portland and many more. View the full list of Dominic Fike tour dates below.
GET TICKETS TO DOMINIC FIKE AT HOLLYWOOD PALLADIUM (NOV. 12)
GET TICKETS TO DOMINIC FIKE AT THE WILTERN (Dec. 17)
GET TICKETS TO ANY DOMINIC FIKE TOUR DATE
Dominic Fike’s upcoming tour will also be his first proper tour run since the release of his debut studio album, What Could Possibly Go Wrong, which was released in 2020. His only musical release do date has been “Elliot’s Song,” a duet with Zendaya for the Euphoria season two soundtrack.
11/06 – Seattle, WA @ Showbox SoDo*
11/08 – Vancouver, BC @ Commodore Ballroom
11/09 – Portland, OR @ Roseland Theater
11/11 – San Francisco, CA @ The Masonic
11/12 – Los Angeles, CA @ Hollywood Palladium
11/16 – Austin, TX @ Stubb’s Waller Creek Amphitheater
11/17 – Dallas, TX @ House of Blues Dallas
11/19 – Houston, TX @ White Oak Music Hall
11/21 – St. Petersburg, FL @ Jannus Live
11/23 – Orlando, FL @ House of Blues Orlando
11/25 – Atlanta, GA @ Tabernacle
11/26 – Nashville, TN @ Brooklyn Bowl
11/28 – Silver Spring, MD @ The Fillmore Silver Spring
11/29 – Philadelphia, PA @ Franklin Music Hall*
12/02 – New York, NY @ Terminal 5*
12/03 – Boston, MA @ House of Blues Boston
12/05 – Montreal, QC @ L’Olympia
12/06 – Toronto, ON @ HISTORY
12/09 – Detroit, MI @ The Fillmore Detroit
12/10 – Chicago, IL @ Radius
12/11 – Minneapolis, MN @ Fillmore Minneapolis
12/13 – Denver, CO @ Ogden Theatre*
12/14 – Salt Lake City, UT @ The Depot
12/16 – Tempe, AZ @ Marquee Theatre
12/17 – Los Angeles, CA @ Wiltern

© 2008-2022 Grimy Goods is first-generation Latina owned & operated since 2008. All rights Reserved
Tiger Talk – Live with Coach Harsin – Auburn Tigers Official Athletic Site
Central mail carrier robbed at gunpoint while working route; master key stolen – WBRZ
CENTRAL – A mail carrier was robbed at gunpoint and had her master key stolen while she was working her route Monday night.
The East Baton Rouge Sheriff’s Office reports the driver was working along the 14400 block of Central Woods Avenue when she noticed an older vehicle following closely behind her. The other vehicle then pulled in front of her and two suspects got out, one of whom took out a gun and took a master dropbox key from the carrier.
No further information was released, and no arrests have been made.
This is a developing story.
Howard University Master of Architecture Program Awarded DoE New Zero Energy Design Designation – The Dig
Bison who are changing the world to make it a better place right now and for future generations.
The U.S. Department of Energy’s (DOE) Office of Energy Efficiency and Renewable Energy awarded its first-ever Zero Energy Design Designation (ZEDD) seal of recognition to Howard University for its master of architecture program with an equitable high-performance energy design concentration.
Earned by only 12 collegiate institutions worldwide, this inaugural designation distinguishes post-secondary academic programs that impart the best practices of zero-energy design on students and require them to apply those building science concepts in actual projects. These leading educational programs are recognized by the DOE for preparing tomorrow’s architectural and engineering leaders to design and build the most sustainable buildings possible.
The ZEDD Program supports the Biden-Harris Administration’s goal of a net-zero emissions economy by 2050. With buildings being one of the main contributors to carbon emissions, building professionals must be trained to design and construct high-efficiency, low-carbon buildings powered by renewables to achieve this goal.
Graduates of these programs will join the front lines of our fight against the climate crisis by designing sustainable buildings that bring the benefits of our clean energy future to all.”
“Our fight against climate change runs straight through our nation’s buildings, and the forward-looking college and university programs we honored today are paving the way for students to lead our net-zero greenhouse gas emissions future,” said Carolyn Snyder, deputy assistant secretary for energy efficiency. “Graduates of these programs will join the front lines of our fight against the climate crisis by designing sustainable buildings that bring the benefits of our clean energy future to all.”
Offered to qualifying programs of study for 3 years, the DOE program requires graduating students to complete a building science education curriculum that uses DOE’s Solar Decathlon Building Science Education learning modules or otherwise meets ZEDD’s learning objectives. Students must also participate in a Zero Energy Design Practicum, either by completing the U.S. Department of Energy Solar Decathlon Design and/or Build Challenge or by engaging in a real-world zero-energy design project that would earn the DOE Zero Energy Ready Home certification or more stringent energy and environmental performance standard.
Our public relations team can connect you with faculty experts and answer questions about Howard University news and events.
Office of University Communications
1851 9th Street NW, 4th Floor, Washington, DC 20059
Phone: 202.806.2509
Fax: 202.238.2340
© 2022 Howard University.
All Rights Reserved.
Native Instruments pads out its Komplete 14 suite with some welcome new toys – Engadget
Komplete, Native Instruments’ flagship music production bundle, has a little bit of everything. That’s always been part of its appeal. It’s pricey, but you get monster synths, a top-notch drum sampler, a virtual guitar rig and Kontakt — which is also a sampler, but calling it one seems incredibly reductive. Native Instruments is still one of the biggest names in the music software world, but it’s an increasingly crowded and competitive market. And much of it is moving towards a subscription model (even Native Instruments). So this year the company is adding some new software in hopes that customers will come back for at least one more big-ticket purchase.
Komplete 14 is the first version to be released since Native Instruments (NI) joined Soundwide, a collection of brands including Izotope and Plugin Alliance, among others. As such, one of the biggest additions to the Komplete library (at least in the $599 Standard version and higher) is Izotope’s Ozone 10 Standard. This mastering plug in has legions of fans thanks to its powerful feature set and simple interface. But for many, the biggest selling point is it’s AI-powered mastering assistance. Many amateur musicians (myself included) rely on Ozone to master their tracks. You simply play the loudest bit of your song, click a button, and the plugin will suggest a starting point for mastering including compression and EQ. You can then accept the settings, tweak them to your liking or toss them and start from scratch.
The new partnership also allowed Native Instruments to beef up the bundle with a handful of smaller items from Plugin Alliance and Brainworx like bx_Oberhausen, bx_Crispytuner and LO-FI-AF. None of these instruments or effects individually are likely to convince you one way or another that Komplete’s more expensive versions are worth the outlay. But I don’t know anyone who is going to complain about having too many plugins.
The only issue is it might not be immediately clear to many users how to get access to those. They’re not in the Native Access manager. Instead you’ll have to go to your products and serials list on the NI site to get the “Plugin Alliance Bundle for Komplete 14” code. Then you’ll head on over to Plugin Alliance and redeem it the code and download a separate plugin manager. Hopefully at some point the two platforms will be integrated to remove the addition steps.
The big centerpiece of Komplete, as always, is Kontakt. The new version — seven — isn’t a giant departure for this industry stalwart. The browser has been updated for better compatibility with HiDPI displays and improved search and filtering tools. The factory library has also been overhauled to take advantage of the graphical overhaul and for better sound. The process of building your own Kontakt instruments has also been simplified with improved creator tools.
Kontakt 7 may not be a significant change from version six, but if you spring for the more expensive versions, like the $1,199 Komplete Ultimate or $1,799 Komplete Collector’s Edition, you do get some unique and powerful expansions like Lores, Ashlight, Kinetic Toys and, one of my personal favorites, Piano Colors. The latter combines samples of a grand piano, various synths and textures, along with effects and modulation tools to create complex sounds that walk the line between organic and synthetic.
The one piece of bad news here is that Kontakt 7, while it is technically included in Komplete 14, isn’t available yet and wont ship until some point in October. Komplete 14 is available now starting at $199 for the basic Komplete Selects package, and going all the way up to $1,799 for Komplete Collector’s Edition.
Subscribe to our two newsletters:
– A weekly roundup of our favorite tech deals
– A daily dose of the news you need
Please enter a valid email address
Please select a newsletter
Navy master chief loses rank and given restriction at court-martial – NavyTimes.com
[Editor’s note: The headline on this story has been changed to accurately reflect his punishment.]
The former senior enlisted leader of a California-based training command was convicted this summer on charges that he sexually harassed other sailors and engaged in sexual activity with junior service members in training, officials confirmed this week.
Cryptologic Technician (Interpretive) Senior Chief Adam I. Shucard “was found guilty of all charges and specifications at a summary court-martial” on June 17, according to Naval Education and Training Command spokeswoman Carla McCarthy.
The former master chief for Information Warfare Training Command Monterey was reduced in rank to E-8 and given 60 days’ restriction, McCarthy said.
Shucard could not be reached for comment and the Navy’s Defense Service Office West did not respond to emailed requests to be put in touch with Shucard’s attorney.
The charges against Shucard stemmed from incidents that occurred after he arrived at the command in July 2020, according to a copy of his charge sheet provided to Navy Times earlier this year.
Shucard was charged with “engaging in conduct that constitutes sexual harassment” from August 2020 to August 2021 and also from September 2020 to August 2021.
He was also charged with abusing a training leadership position for allegedly engaging in sexual activity with a sailor from August 2020 to August 2021, “while in a position of authority” over the sailor, a shipmate whom Shucard “knew or reasonably should have known was a specially protected junior Servicemember in initial active duty training,” according to the charge sheet.
He faced another specification of the abusing a training leadership position charge for engaging in sexual activity with a sailor over whom he had authority from September 2020 to August 2021, the charge sheet states.
Shucard is also charged with four specifications of failure to obey a lawful general regulation for “engaging in a personal relationship” that was “unduly familiar and that did not respect differences in grade or rank” over several time periods from August 2020 to August 2021, according to his charge sheet.
A California native, Shucard enlisted in 1994 and graduated from the Defense Language Institute’s Persian program before undertaking information warfare and information operations assignments worldwide, according to his command biography.
Geoff is a senior staff reporter for Military Times, focusing on the Navy. He covered Iraq and Afghanistan extensively and was most recently a reporter at the Chicago Tribune. He welcomes any and all kinds of tips at [email protected].
Navy Times © 2022
Navy Times © 2022
PRESONUS RELEASES STUDIO ONE™ 6, COMPREHENSIVE UPGRADE TO AWARD-WINNING DAW, UNLOCKING CREATIVITY FOR COMMITTED PROFESSIONALS ACROSS GENRES – Yahoo Finance
Best-In-Class Music Production Software Adds Industry Firsts, Customizable Interface, Lyric Support, Optimized Mixing Tools, Global Video Track, Smart Templates, New Plugins, Collaboration Enhancements and More
HOLLYWOOD, Calif., Sept. 29, 2022 /PRNewswire/ — PreSonus Audio Electronics, Inc. today announced the global launch of Studio One™ 6, the latest version of the company’s award-winning music production software for macOS and Windows. Studio One™ 6 is the most recent chapter in the evolution of one of the bestselling digital audio workstations in the world, featuring industry-leading innovations, mixing improvements and the most advanced user interface customization options available. Version 6 makes audio content creation fast and easy, with everything you need to record, produce, mix, master and perform in a single intuitive application. Studio One 6 is available for purchase as a perpetual one-time license, or as an annual or monthly subscription to PreSonus Sphere via PreSonus.com and participating dealers. In addition to the latest two versions of Studio One™ Professional, PreSonus Sphere™ members get access to expanded collaboration tools in Studio One™ and online, Community, exclusive educational content, all PreSonus software and add-ons giving the best value to creators and the ultimate Studio One™ experience. With additional landmark innovations in the works for future iterations, Studio One™ 6 is just the start of the rollout of a new and improved feature set to support recording professionals.
“In this latest iteration, the team stayed focused on listening to loyal users’ needs, creating innovative solutions to boost productivity in the professional audio production space,” said Arnd Kaiser, General Manager, Software at PreSonus. “We want engineers, producers, recording musicians and creators around the world to feel supported and inspired to build their best work yet. Studio One™ has evolved over the past 12 years and is now an established major player in the audio production space known for unique features and a fast, intuitive workflow. We continue to push the boundaries of what’s possible in recording and are now the first digital audio workstation in the industry to feature end-to-end lyric support for both audio and MIDI, from songwriting to recording and live performance.”
“We are excited for the launch of Studio One 6 and are fully invested in supporting the continued expansion and adoption of Studio One as the total solution for professional and committed at-home creators,” said Evan Jones, CMO, FMIC. “The global adoption of digital creative and productivity tools continues to grow and expand across all forms of musical expression, in the hands of an even wider audience of creators, producers, mix engineers, and artists. Together with PreSonus hardware and Studio One™ we’re providing the total solution for professional and committed at-home creators, and we can’t wait to hear the music they’ll make.”
Studio One™ 6 offers major new features and enhancements to boost creativity for seasoned professionals; moreover, the redesigned and customizable user interface helps creators focus on the task at hand. Studio One™ 6 is available in both perpetual (priced at $399.95 USD for Professional) or as part of the PreSonus Sphere™ subscription, which gives members every upgrade for free. Going forward, PreSonus Sphere™ members ($14.95/month or $164.95 annual) will always have access to the current version of Studio One™ Professional as well as one version back, giving them the freedom to upgrade whenever it works best with their projects.
Studio One™ 6’s top new features include:
Smart Templates: New Smart Templates combine preconfigured tracks and channels with a customized user interface and automatic download of required content to get started recording and producing right away. Smart Templates provide essential tool sets for the tasks at hand, including basic recording, beat making, mixing stems, mastering an album, setting up for a live show and more. Smart Templates include a convenient drop zone to easily import needed content like audio, video and MIDI files as well as interactive tutorials with instructions and links to related video content.
Customization Editor: The new Customization Editor provides a convenient and easy way to show or hide nearly all tools and functions, customizing the user interface to create a streamlined, focused workflow for the task at hand. First-time users, seasoned professionals and everyone in between can create a Studio One™ interface customized for their needs. Best of all, customizations can be saved to recall at will — a few for essential workflows are provided as jumping-off points.
Industry-leading Lyrics Support: Studio One™ 6 now provides powerful, industry-leading lyric support. Lyrics can be added to scores, MIDI productions, audio sessions – even live performances! Use the all-new Lyrics Display as an editor, a teleprompter or a quick, innovative way to navigate a lyric-laden production. When working with Score or Piano view, lyrics can be attached to notes word by word or by syllable. Lyrics can also be transferred to the Show Page to use during rehearsals or performances.
Global Video Track: Content creators and sound-to-picture professionals alike will appreciate the all-new Global Video Track and its intuitive workflow. Add video clips by simply dragging video files onto the timeline. The Video Track can hold an unlimited number of video clips, making it easy to work on multiple reels of a feature film in a single session. Video thumbnails are generated automatically to provide a preview for easy navigation. A dedicated sub-track is provided for embedded audio to use as reference, complete with mixer channel, full plug-in processing and output routing options.
Advanced PreSonus Sphere™ Collaboration: Studio One™ 6 deepens the integration with the PreSonus Sphere™ fully loaded with powerful, cloud-based collaboration Workspaces, using convenient sharing options and built-in sync options right from the main menu. An all-new comprehensive installation window lets members quickly explore, download and install exclusive PreSonus Sphere™ content, with a variety of display, search and filter options.
New Plug-ins: It wouldn’t be a Studio One™ upgrade without new and improved Native plug-ins and version 6 is no exception. The Studio One™ new De-Esser is an efficient and intuitive tool for taming vocal sibilance and other high-frequency sounds. The new Vocoder is a creative playground that lets users turn a voice or any audio source into a unique synth or effect. Finally, the ever-popular ProEQ, Autofilter, Mai Tai, Sample One XT and Presence XT have all been significantly enhanced and improved.
“From 2017 on, Studio One™ has been the official DAW here at Studio DMI, in every room, for every purpose,” said Luca Pretolesi, Grammy Nominated Mixing and Mastering Engineer at Studio DMI Las Vegas. “Every record I’ve worked on so far, from big album projects like Drake to dance records for Diplo, has been done in Studio One™. PreSonus Studio One™ 6 is out and it’s great. There are so many improvements and I think everyone at PreSonus read my mind. They elevated certain functionality and features that are very important for my workflow, especially the Channel Overview. Staying on one channel and being able to access everything on the channel is a key element for me. I love the additional function on the sidechain; it was already crazy before. The ProEQ3 is even better. The dynamic functionality and the solo make it the ultimate EQ in a plugin format.”
Studio One™ 6 introduces a long list of additional new features and enhancements, including improvements to the Console, the Browser, the Start Page and much more. In addition, the Studio One™ 6 Console is also packed with new features developed at the request of the loyal Studio One™ community. Fader Flip lets users control FX/Bus sends, sidechains and Cue Mix sends all from the main faders. Stereo channels now feature a choice between balance, dual or binaural panning as well as a new panner pop-up editor. The new Channel Overview provides a single, consolidated view of all channel parameters, including input controls, inserts, sends and Cue Mix sends for the selected channel that can be pinned to keep a critical eye on individual channels. Finally, Studio One™ 6 takes workflow efficiency to a new level with Track Presets, storing every aspect of a given track or channel for instant recall, even for multiple selected tracks/channels.
Watch the official launch video on Studio One™ 6 HERE. High-resolution product images can be found HERE. Full product descriptions and specifications can be found HERE.
For technical specs, additional information on new PreSonus products and to find a retail partner near you, visit www.presonus.com and www.fender.com. Join the conversation on social media by following @PreSonus.and @Fender.
ABOUT STUDIO ONE™ 6: The latest version of the PreSonus® award-winning, best-selling music production software for macOS and Windows, Studio One™ 6 makes creating fast and easy with everything you need to record, produce, mix, master and perform all in a single intuitive application. Version 6 adds a wealth of new features and enhancements, including a customizable interface that lets you focus on the task at hand, a new Global Lyrics Track and Global Video Track, deeper integration with PreSonus Sphere™ Workspaces, new and improved Native plug-ins and much more. Studio One™ 6 makes content creation easier for first-time artists and seasoned professionals alike.
Pricing is as follows:
PreSonus Sphere™: ($14.95/mo., $164.95/yr. USD or €12.20/€134.10 EUR)
Studio One™ 6 Professional digital audio workstation for macOS and Windows: $399.95 USD
Studio One™ 6 Artist: $99.95 USD
All previous Studio One™ Professional versions to Studio One™ 6 Professional Upgrade: $149.95 USD
All previous Studio One™ Artist versions to Studio One™ 6 Professional Upgrade: $299.95 USD
All previous Studio One™ Artist versions to Studio One™ 6 Artist Upgrade: $49.95 USD
Studio One™ Professional 30-Day Demo: Free* & Studio One™ Prime: Free
ABOUT PRESONUS AUDIO ELECTRONICS, INC.:
PreSonus Audio Electronics Inc. is the leading designer and manufacturer of recording and live-sound hardware and software solutions that solve real-world problems for creatives of all levels. Founded in 1995 and headquartered in Baton Rouge, La., PreSonus designs innovative audio tools for musicians, content creators, producers and audio engineers, providing professional sound quality and features without sacrificing affordability. Today, PreSonus is an owned brand under Fender Musical Instruments Corporation (FMIC). While most guitar and bass players still plug into amplifiers, many also plug their instruments into interfaces, recording songs as well as using virtual amps and effects to create their sounds. With a shared passion for music and a commitment to innovation, FMIC and PreSonus are focused on delivering an ecosystem that seamlessly integrates hardware and software for an effortless end-to-end experience. For more information, visit www.presonus.com.
ABOUT FENDER MUSICAL INSTRUMENTS CORPORATION:
Since 1946, Fender has revolutionized music and culture as one of the world’s leading musical instrument manufacturers, marketers and distributors. Fender Musical Instruments Corporation (FMIC) — whose portfolio of owned and licensed brands includes Fender, Squier, Gretsch guitars, Jackson, EVH, Charvel, Bigsby and PreSonus — follows a player-centric approach to crafting the highest-quality instruments and digital experiences across genres. Since 2015, Fender’s digital arm has introduced a new ecosystem of products and interactive experiences to accompany players at every stage of their musical journey. This includes innovative apps and learning platforms designed to complement Fender guitars, amplifiers, effects pedals, accessories and pro-audio gear, and inspire players through an immersive musical experience. FMIC is dedicated to unlocking the power of musical expression for all players, from beginners to history-making legends. In 2021, Fender celebrated 75 years of giving artists “wings to fly,” carrying on the vision of its founder, Leo Fender, and connecting players through a shared love of music.
PreSonus, Studio One (word and logo) and PreSonus Sphere are trademarks of PreSonus Audio Electronics, Inc., and/or its affiliates, some of which are registered in the U.S. and other countries.
All other product and company names may be trademarks of their respective owners, and may be used herein under license. Use of them does not imply any affiliation with or endorsement by or of the respective owners.
View original content to download multimedia:https://www.prnewswire.com/news-releases/presonus-releases-studio-one-6-comprehensive-upgrade-to-award-winning-daw-unlocking-creativity-for-committed-professionals-across-genres-301636222.html
SOURCE PreSonus Audio Electronics; Fender Musical Instruments Corporation
Elon Musk is one of the most influential personalities on social media. The billionaire CEO of Tesla has understood that it is on these platforms that trends are made today and opinions are formed. For a large section of consumers and people, social networks are even the main source of information.
Executive claimed remark was a joke
These Apple suppliers are at risk if the tech giant pulls back on iPhone production, warns Bank of America.
The death of Molly Russell "should send shockwaves through Silicon Valley", the NSPCC's chief says.
The skate video has been a staple of the scene since the '80s, but now it's more important than ever.
The inquest into the death of British schoolgirl, Molly Russell, has concluded that social media was a factor in her demise, the BBC reports. The 14-year-old had viewed thousands of pieces of content about self-harm and suicide on online platforms, including Instagram and Pinterest, prior to her death in November 2017. Reaching a conclusion on the North London Coroner's Court inquest into Russell's death today, coroner Andrew Walker said the "negative effects of online content" were a factor in her death and such content "shouldn't have been available for a child to see".
One of Apple's most senior executives has been sacked after he jokingly quoted a Dudley Moore line about being able to “fondle big breasted women” on the job.
The schoolgirl who took her own life aged 14 had a secret Twitter account which reveals her thoughts.
The provider of electric vehicle charging networks could see its shares trade sideways for a bit longer.
A key part of Gloucester City Council's online planning system has been restored, it says.
"Baywatch" star Donna D'Errico took to Instagram on Wednesday to share a photo in a gold, string bikini as she deals with the heat wave in Los Angeles.
Trevor Noah is leaving Comedy Central’s “The Daily Show” after seven years. The South African comedian made the announcement at the end of the taping of Thursday’s show, after thanking the audience for their support. Noah took over hosting the satirical news show from Jon Stewart in 2015.
Zoom Video Communications (ZM) brings new features on Zoom One, Zoom Contact Centre, Zoom Spaces, Zoom Events and Zoom IQ for Sales.
The purveyors of web3 want a slice of this gargantuan market. Back in July, I wrote that veterans from the gaming industry overwhelmingly agree blockchain games should be fun to play and offer a sustainable financial model, both of which are missing in play-to-earn games like Axie Infinity. When it comes to genres, many of them believe massively multiplayer online (MMO) games have the chance to onboard the masses into web3.
A series of text messages released as part of ongoing litigation over the failed Twitter deal reveal the billionaire's plans for a social media platform that would charge users to put short messages on a blockchain.
The new iPhone 14 and Apple Watch models promise to detect severe car crashes. According to our tests, speed and impact are only part of the story.
The cause of Queen Elizabeth’s death has been revealed.
Whitworth Production of "The Sound of Music" | Arts | khq.com – KHQ Right Now
Partly to mostly cloudy. High 69F. Winds light and variable..
Clear skies. Low 49F. Winds light and variable.
Updated: September 30, 2022 @ 2:54 am
Something truly special is happening at Whitworth this fall. The fall main-stage production, The Sound of Music, Oct. 14-23 in Cowles Auditorium, will be a collaboration by Whitworth’s theatre and music departments, with both alumni and current students taking part.
Check out this amazing production, which will run from Oct. 14-23 in the Cowles Auditorium.
Get up-to-the-minute news sent straight to your device.
Fuzzy-clustering and fuzzy network based interpretable fuzzy model for prediction | 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: 16279 (2022)
Metrics details
Interpretability is the dominant feature of a fuzzy model in security-oriented fields. Traditionally fuzzy models based on expert knowledge have obtained well interpretation innately but imprecisely. Numerical data based fuzzy models perform well in precision but not necessarily in interpretation. To utilize the expert knowledge and numerical data in a fuzzy model synchronously, this paper proposed a hybrid fuzzy c-means (FCM) clustering algorithm and Fuzzy Network (FN) method-based model for prediction. The Mamdani rule-based structure of the proposed model is identified based on FCM algorithm from data and by expert-system method from expert knowledge, both of which are combined by FN method. Particle swarm optimization (PSO) algorithm is utilized to optimize the fuzzy set parameters. We tested the proposed model on 6 real datasets comparing the results with the ones obtained by using FCM algorithm. The results showed that our model performed best in interpretability, transparency, and accuracy.
Fuzzy model is to implement logical reasoning and intelligent calculation for fuzzy information without a definite mathematical model. The main method is a set of IF–THEN rules-based procession designed either from expert knowledge or numerical data. Although independent on accurate mathematical model, fuzzy model is a powerful technique for logical reasoning, numerical calculation and non-linear function approximation1.
To identify a fuzzy model, structure establishment and parameter estimations are two main steps2. Structure identification is related to the number of rules after the important input variables have been selected. Parameter estimation (fuzzy set parameters) descripts a reliable non-linear approximation system.
Interpretability is the major feature of a fuzzy model in special fields, for example in security-oriented fields, and military field. There is a large amount of knowledge on top of the experience of experts. When this kind of knowledge is dealt with by machine learning, the security of this fuzzy logic (FL) model is vital for military application, and also fundamental for the transparency and interpretability of the knowledge. This kind of knowledge expressed by FL is conceptually easy to understand, tolerant of the imprecise information, and easy for human communication. Thus, expert knowledge based fuzzy models are traditionally well interpretable3, which is the motivation of the approach with fuzzy logic.
Expert knowledge is mainly expressed in Mamdani-type IF–THEN rules because this structure is more interpretable than T-S-type ones. Arguably, traditionally fuzzy models based on expert knowledge have obtained well interpretable innately. So, Mamdani rule-based structure is suitable for an interpretable-oriented fuzzy model.
Numerical data based fuzzy models are not necessarily interpretable, but lots of them perform well in precision with the help of soft computing techniques. For example, artificial Neural network (ANN), inspired by neuroscience, is one of the most successful methods in “learning” from data. Modern ANNs, particularly “deep learning” models, have been sped up by the increase of raw computer power. But “they cannot approach the cognitive capabilities of a four-year old. Perhaps more striking is that ANNs remain even further from approaching the abilities of simple animals4.” It implies that the innate structure plays the dominant role in the learning capability.
The FCM clustering algorithm is a powerful unsupervised learning technique to form a few rules with simple and interpretable structure5. FCM induces rules by organizing and categorizing data into partitions. Partitions with homogeneous data form clusters, and each cluster is associated with a rule. The fuzzy sets of rules are independent from each other. Each dimension of data is tailored only for one rule6. In order to improve the precision, many optimization-based FCM models have been proposed combining with metaheuristic optimization algorithms, such as genetic algorithm(GA)7,8 and PSO9,10,11,12,13,14,15. Both GA and PSO aim to solve optimization problems without being trapped into local minima. Due to its versatility and simplicity, PSO has become one of the most popular metaheuristics and an important tool for many applications.
The fuzzy models for prediction should utilize the expert knowledge and numerical data in a fuzzy model synchronously16,17. Historical data provides numerical quantitative measurements from past projects regarding the internal and external quality attributes. Experts utilize their experience to provide “fuzzy” information, or qualitative descriptions of the correlation between the internal and external quality attributes18. To make the model for prediction more practical, interpretability is crucial as it allows the practitioners to provide their own judgment on the predictors in linguistic terms. Black box prediction models (e.g., ANN-based models), are hard to identify the structure and to incorporate the experts’ judgments.
The novel method of FN is capable to combine the rules both from knowledge and from data. As a kind of Chained Fuzzy System (CFS)19 or Hierarchical Fuzzy System (HFS)20,21, FN maps the inputs to the outputs by means of connections22,23,24. The overall number of rules in FN is a linear function of the number of inputs and the number of linguistic terms per input. Compared to Standard Fuzzy System (SFS), the rules in FN are reduced and simplified. The structure in FN is more transparent and interpretable. Arguably, FN is characterized by interpretable innately as a white-box model. The details about FN will be introduced in section “Related methods”.
This paper proposes a hybrid FCM and FN based model for prediction (FCM-FN). FCM-FN is Interpretability-oriented model with Mamdani rule-based structure. The rules are generated by FCM method from data in the first step, and then by expert-system method from expert knowledge in the second step. Afterwards, the rules in both previous steps are connected by FN method. The Interpretable structure of FCM-FN model is identified. As a kind of prediction model, FCM-FN pursues accuracy while preserving interpretable structure. PSO algorithm is utilized to optimize the fuzzy set parameters which is initialized by FCM algorithm in the first step, and by expert system in the second step. For simplicity, the FCM-FN is a kind of multi-input single-output (MISO) type-1 fuzzy system. A multi-input multi-output (MIMO) fuzzy system can be taken as the composition of several MISO fuzzy systems25. As for type-2 fuzzy system, the related theories and practices have achieved diverse developments26, but the interpretability is rarely included into the type-2 theory so far. This is the reason why the FCM-FN is limited in type-1 structure in this paper.
The rest of the paper is synthesized as follows: section “Related methods” describes the Mamdani’s fuzzy model, FCM method, PSO method and FN theory, which will be used in this paper. In section “The proposed fuzzy model for prediction” the structure of the proposed FCM-FN model is analyzed in detail. The effectiveness of the model is illustrated in section “Case study”, through applications to various real datasets. The paper ends with the concluding remarks in section “Conclusion”, where the proposed approach is summarized, and its main characteristics are identified.
In this section, the related methods mentioned in the previous section are now further explained. They are divided in five parts: Mamdani-type fuzzy inference, FCM, PSO, PSO based tuning membership functions, and FN.
Inspired by Zadeh27,28, one of the most interpretable fuzzy models was suggested29, in which Mamdani attempted to control a steam engine and boiler combination by synthesizing a set of linguistic rules from experienced human operators (expert knowledge). The literature on the Mamdani-type fuzzy logic has grown rapidly. One of them was implemented as an universal approximator17. Mamdani-type fuzzy inference includes four steps. In order to explain Mamdani-type fuzzy inference, the example17 is employed where the M rules are formed as Eq. (1).
where j = 1,2,…,M, xi(i = 1,2,…,n) are the input variables to the fuzzy system, z is the output variable, (A_{i}^{j}) and Bj are linguistic terms of the linguistic variables xi and z in the universes of discourse U and R, respectively.
The first step is to take the crisp inputs, and determine the degree to which these inputs belong to each of the appropriate fuzzy sets. Membership functions associating weighted inputs define functional overlaps between inputs, and ultimately determine output responses. Because membership functions are graphical representations of the magnitude of input participations in fuzzy logic, a fuzzy set is defined by its membership functions. Let X be a set of items, known as the universe, and its elements are denoted by x. And, a fuzzy subset A in X is characterized by the membership function μA(x) which is associated with each element x in A and a real number in the interval [0, 1]. The membership function μA(x) maps each element x to a membership value, which represents the level of membership of x in A. Different membership functions can be associated with different inputs and outputs. In essence, they are weighting factors for the outcomes of fuzzy rules. Gaussian or triangular shape are two well-known membership functions.
Gaussian membership function is specified by two parameters as Eq. (2).
where θ is the position of the peak relative to the universe, σ is the standard deviation.
Symmetric triangular membership function is also specified by two parameters as Eq. (3) and Fig. 1.
Symmetric triangular membership function.
The fuzzified inputs are applied to the rule antecedents. As the fuzzy rule has multiple antecedents, the AND or OR fuzzy operation is used to obtain a single number which represents the result of the antecedent evaluation. In this paper, the AND is applied to evaluate the conjunction of the rule antecedents, which is defined as Eq. (4).
In other application, if the OR fuzzy operation applies the classical fuzzy operation union, which is defined as Eq. (5).
Now the antecedent evaluation result can be used to the membership function of the consequent, which is commonly based on two methods: clipping and scaling. Clipping method is to cut the consequent membership function by the truth value of the rule antecedent, and scaling method is to adjust the original membership function of the rule consequent by multiplying all its membership degrees at the level of the antecedent truth. Clipping loses some information, but involving less complex and easier to defuzzify. And scaling preserves the original shape of the fuzzy set. For simplification and interpretation, the method clipping is adopted in this paper.
The membership functions of all consequents clipped in the previous step are combined into a single fuzzy set.
The centroid technique is a popular defuzzification method. It finds the centroid point representing the center of gravity (COG) of the aggregated fuzzy set R. A reasonable estimate can be obtained by centroid deffuzification method, which is defined as Eq. (6).
where (overline{z}_{{^{{^{j} }} }}) is the point in R at which (mu_{{B^{j} }} (z)) achieves its maximum value (usually, we assume that (mu_{{B^{j} }} left( {overline{z}_{{^{{^{j} }} }} } right)) = 1).
This method is a universal approximator, i.e. they can approximate any continuous function on a compact set to an arbitrary accuracy17.
Clustering is an unsupervised learning method, which assigns a given set of objects into disjoint groups or clusters by membership degrees between 0 and 1. A high degree value represents a high similarity between the object and the group. FCM is a well-known fuzzy clustering algorithms30. The main motivation of proposing FCM was to address the deficiency in working with overlapping groups shown by the hard cluster algorithm. The FCM method was developed by researches recently31,32,33,34. In order to keep the simplicity and interpretability, the original FCM method is included in FCM-FN model. The process is expressed as follows9:
Let (Omega = { 1, ldots ,k, ldots ,n}) be a set with n objects. Object k is the vector of quantitative variables (X_{k} = { x_{1k}, ldots ,x_{jk}, ldots ,x_{pk}}) described by p variables where (x_{jk} in {varvec{R}}). Let (Y = { 1, ldots ,i, ldots ,c}) be a set of c prototypes associated to c groups, where each prototype i is a vector of quantitative variables (Y_{i} = { y_{1i}, ldots ,y_{ji}, ldots ,y_{pi}}), where (y_{ji} in {varvec{R}}). Let (U = left[ {u_{ik}} right]) be a c × n membership degree matrix, where ({text{u}}_{ik}) is the membership degree of object k to group i, where ({text{u}}_{ik} in [0,1]). The algorithm works according to minimizing the objective function that is defined as Eq.(7). At the same time, a prototype matrix (Y^{*}) and a membership degree matrix (U^{*}) are obtained.
where m is the value of cluster fuzziness, and (d_{ik}) is the squared Euclidean distance which measures the dissimilarity of the feature vectors between (x_{k}) and (y_{i}). The distance is calculated by Eq. (8).
Under the minimizing criterion J, the prototypes are updated according to Eq. (9), and the membership degrees are updated by Eq. (10) with the restriction (sumlimits_{i = 1}^{c} {u_{ik} = 1}).
As mentioned in section “Introduction”, PSO is a typical optimization algorithm that searches for the best solution by modeling the social behavior of bird flocks and fish schools. The population of the birds is named “swarm”, and the members of the population are called “particles”. Assume that the Search space dimension is D, and the swarm is Sw = ({ X_{1} , ldots ,X_{s} }), where s is the total number of particles; A particle represents a position in D; the position of the ith particle in the search space is denoted as (X_{i} = left( {x_{i1} , ldots ,x_{id} , ldots ,x_{iD} } right),;) i = 1,2,…,s. pbest records the position of ith particle’s previous optimal performance, which is denoted as (p_{i} = left( {p_{i1} , ldots ,p_{id} , ldots ,p_{iD} } right)). gbest is the best position achieved by the swarm, which is expressed as (p_{g} = left( {p_{g1} , ldots ,p_{gd} , ldots ,p_{gD} } right)). The velocity of the ith particle is denoted as (v_{i} = left( {v_{i1} , ldots ,v_{id} , ldots ,v_{iD} } right)). pbest and gbest direct a particle’s new velocity and position. (p_{id}^{{^{{}} }} (t)) is the ith particle’s optimal position, and (p_{gd}^{{}} (t)) is the global best position of the dth dimension at instant t. The particles fly through the solution space towards better positions, and the process is implemented by Eqs. (11) and (12).
where the positive constants (c_{1}) and (c_{2}) are acceleration coefficients, r1 and r2 are random values in range [0,1]. (x_{id} {text{(t)}}) is the position and (v_{id} (t)) is the velocity of ith particle in dth dimension at instant t. The inertia value (omega (t)) is obtained by Eq. (13).
where t is the current instant and (t_{max }) is the maximum iteration number. When iteration proceeds, the weighting factor of updating speed will decrease from the maximum factor (omega_{max }) to the minimum one (omega_{min })11,15.
To solve clustering problems, each particle represents a feasible solution to the optimization problem. Let f is an object function, and the personal best position of a particle at instant t is updated by Eq. (14).
where (p_{g}^{{}} (t + 1)) is the global best position at instant t+1 found by anyone of all particles during the previous steps, defined by Eq. (15).
In the proposed model, there are two kind of membership functions: Gaussian or triangular. PSO algorithm is specified to adjust the shapes of membership functions by tuning the parameters of membership functions.
The first tuning is aiming towards the parameters of Gaussian membership functions, which are generated by FCM. This process is abbreviated as FCM-PSO. The steps are described by Algorithm FCM-PSO(Ω, c), which returns matrix Y including two kinds of parameters(θ and σ) in Gaussian membership functions, as shown in Eq. (2).
The second tuning is aiming towards the parameters of symmetric triangular membership functions, which is specified by experts. This process, similar to FCM-PSO, returns matrix M including two kinds of parameters(a and α) in symmetric triangular membership functions, as shown in Eq. (3) and Fig. 1.
In this section, the details of the novel theory of FN are expressed, which include the basic theory, the basic operations and three important theorems. All the selected details are involved in the proposed model.
Standard Fuzzy System (SFS) is the most known type of fuzzy system which is with a single rule base. SFS is characterized by the nature of a black-box where the outputs map the inputs directly, and internal connections are out of consideration35,36. Reflecting the influence of all inputs on the output simultaneously, SFS is quite accurate for output modeling usually. But, when the rules of SFS increasing, the transparency deteriorates, and it is less clear how the inputs affect the output.
Chained Fuzzy System (CFS), is with multiple rule bases in cascade, and characterized by the nature of white-box where the outputs map the inputs by connected internal variables37,38. CFS is with an arbitrary structure with the form of connected subsystems. It is applied as a detailed presentation of SFS for improving transparency by explicitly accounting all subsystems’ interactions. And the smaller number of inputs improve the efficiency of CFS. However, because of the error accumulation from the multiple Fuzzification-Inference-Defuzzification (FID) sequences, accuracy may be lost. Hierarchical Fuzzy System (HFS), a special kind of CFS, each subsystem in HFS only with one output and two inputs.
FN, a novel concept with networked rule bases18,22,24, is characterized by the nature of a white-box where the outputs map the inputs by connections. Arguably, FN is a hybrid of SFS and HFS. On one hand, the structure is similar between FN and HFS because of the explicit presentation of the interactions and subsystems. On the other hand, after the multiple-rule-bases of FN been simplified to a linguistically equivalent single rule base, the operation of FN is similar to SFS. The process of simplification implements the linguistic composition approach, including the vertical merging and horizontal merging of rule bases in FN. The multiple rule base system in HFS is converted into a FN, and then a FN is composed into a single-rule-base system.
As a hybrid concept, FN obtains the advantages of both accuracy from SFS and transparency from CFS/HFS. The structure’s transparency is directly related to interpretation of a prediction model, which has turned out to be same important than accuracy and efficiency for complex systems modelling24, and it is the core reason that FN is employed in the fuzzy model for prediction in this paper.
Inputs and outputs of a FN model take linguistic terms. In this paper, the FN model is with only one output(MISO) and the If–then rules are in form of Eq. (1). For compactness, the linguistic terms of the inputs (left( {left{ {A_{1}^{j} , ldots ,A_{n}^{j} } right}} right)) and the outputs (left( {left{ {B^{j} } right}} right)) in Eq. (1) are represented by positive integers. For example, {‘small’, ‘average’, ‘big’} are encoded as positive integers {‘1’, ‘2’, ‘3’}.
In order to illustrate the basic operations in FN, a simple example is shown as Fig. 2. A FN with 3 nodes N11, N12, I21 is described by the Boolean matrices given in Table 1. For simplicity, Each input (output) is only with two linguistic terms, represented by {‘1’, ‘2’}. More examples have been included in the folder “Mathlab functions” in the Supplementary File.
Hierarchical structure example for illustrating the basic operations in FN.
The rules of N11 are same to the if–then rules given in Eq. (16). The node of N12 is similar to N11.
The special node of I21 is an identity node, and the outputs is same to the inputs. The rules of I21 is described by if–then rules given in Eq. (17).
Vertical merging is a kind of binary operation that can be applied to a pair of parallel nodes, i.e. nodes locate in the same layer of a FN. In Fig. 2, the nodes N11 and I21 are in the same layer. The vertical merging operation is identical with Boolean matrix Kroneker product. This kind of operation merges the operand nodes from the pair into a single product node. In this case, the inputs to the product node represent the union of the inputs to the operand nodes where the outputs from the product node represent the union of the outputs from the operand nodes. The operation of vertical merging can always be applied due to the ability to concatenate the inputs and the outputs of any two parallel nodes24. The symbol ‘+’ denotes the vertical merging operation.
In Fig. 3, nodes N11 and I21 represent a two-node subnetwork of this FN. This two-node FN can be described by the block-scheme and the topological expression in Eq. (18), The vertical merging of the operand nodes N11 and I21 results into a single product node N11+21 which represents a simplified image of the two-node FN in the form of a one-node FN. Due to the concatenation of the inputs to the operand nodes as inputs x1, x2 and x3 to the product node and the concatenation of the outputs from the operand nodes as outputs z1 and x3 from the product node. This node N11+21 can be described by the topological expression in Eq. (19) and the Boolean matrix in Table 2.
Block-scheme of two nodes’ vertical merging into one node with 3 inputs and 2 outputs.
Each node in Eqs. (18) and (19) is placed in a pair of square brackets ‘[ ]’. The inputs and the outputs for each node are placed in a pair of simple brackets ‘( )’ right after the node, where the outputs are separated from the inputs by a vertical slash ‘|’. The topological expressions in the rest of this paper are with the similar structure.
Horizontal merging is a binary operation that can be applied to a pair of sequential nodes, i.e. nodes located in the same level of a FN. This operation merges the operand nodes from the pair into a single product node. The product node has the same input as the input to the first operand node and the same output as the output from the second operand node whereas the connection does not appear in the product node.
As described by the block-scheme in Fig. 4, the two-node FN can be expressed by the topological Eq.(20). The symbol ‘*’ implies that the operation of horizontal merging. The horizontal merging of the operand nodes N11+21 and N12 results into a single product node N(11+21)*12. As a simplified image of the two-node FN in the form of one-node, N(11+21)*12 is descried by the topological expression in Eq. (21) and the Boolean Matrix in Table 3.
Block-scheme of two nodes’ horizontal merging into one node with 3 inputs and 1 outputs.
The three theorems are the important theoretical base for the proposed model in this paper22.
Denoted by the symbol ‘+’, the vertical merging operation is associative according to Eq. (22).
where for any three operand nodes M, N, O, the vertical merging from bottom to top is equal to their vertical merging from top to bottom.
Denoted by the symbol ‘∗’, the operation of horizontal merging is associative according to Eq. (23).
where for any three operand nodes M, N, O, the vertical merging from right to left is equal to their vertical merging from left to right.
A HFS with inputs ({ x_{1} ,x_{2} , ldots ,x_{{text{m}}} }), network nodes ({ N_{11} ,N_{12} , ldots ,N_{{text{1,m – 1}}} }), connections ({ z_{1} ,z_{2} , ldots ,z_{{text{m – 2}}} }) and a output y, as described by the topological expression in Eq. (24)
can be characterized as a SFS with the same set of m inputs, the same single output, a single network node N, and no connections as described by the block-schemes in Fig. 5 and the topological expression in Eq. (25).
where ({ }N = mathop prod limits_{p = 1}^{m – 1} left( {N_{1p} + mathop sum limits_{q = p + 1}^{m – 1} I_{pq} } right)), node Iqp is identity node in level p and layer q.
Block-scheme of Fuzzy Networks for theorem expression.
This section describes the proposed FCM-FN model, the process of construction the model includes 4 steps, as shown in Fig. 6. The first step aims at selection of input variables, based on the available numerical data and expert knowledge. The second step aims towards generation of knowledge base. The semantic rules and their data bases (DB1) are constructed from expert knowledge, and the clustering rules and its data base (DB2) are generated from numerical data by FCM algorithm. Then, the parameters of DB2 are optimized by the PSO algorithm. In the third step, the two kind of rule bases in second step are combined based on FN method. The forth step aims to optimize the parameters of DB1 also based on PSO algorithm. After the four steps, the proposed model as Fig. 7 is converted into a SFS.
Flow sheet of construction the FCM-FN model.
FN structure of the proposed model.
Being converted into a SFS, the proposed FCM-FN model is a MISO system consisting of four major modules: fuzzification, inference engine, defuzzification and knowledge base. The fuzzification module converts the crisp inputs of independent variables to linguistic terms. And these linguistic terms combined with the ones from experts are processed in fuzzy domain by inference engine based on FN. The knowledge base is composed of the rule base, characterizing the control goals and control policy by a set of rules; and of the data base, containing the linguistic term sets and the membership functions defining their semantics. Finally, the processed output is transformed from fuzzy domain (linguistic terms) to crisp domain (numerical number) by defuzzification module18.
There is large amount of indices to character a product. To reduce the complexity of a model, only those metrics and attributes with significant contribution to the prediction are selected. In practices, experts always are with the knowledge that some metrics are more important than others. Pearson Correlation Coefficient ((r_{p}), expressed as Eq. (26)) and Spearman Correlation Coefficient ((r_{r}), expressed as Eq. (27)) indicate the strength and direction of a linear relationship between two variables, which are useful to get top relevant input variables.
where x is values in the first set of data, y is values in the second set of data, n is total number of values.
where (d_{i}^{2}) is the difference between the x rank and the y rank for each pair of data, (sum {d_{i}^{2} }) is sum of the squared differences between x and y variable ranks, n is sample size.
As showed in Fig. 7, the selected metrics xi (i = 1, 2,…, m + n) are input variables. there are m + n inputs divided into two sets. SET1 is composed of numerical inputs xi (i = 1, 2,…, m) which are from the variables with numerical data, and SET2 is composed of semantic inputs xi (i = m + 1, m + 2, …, m + n) from the variables only with the expert’s linguistic information. Combined with the numerical data of output z1 (here z1 is the prediction variable), the numerical data are transformed into a set of input–output data pairs.
The process of generation and optimization of knowledge base includes three parts: generation of rule base, generation of data base and parameters optimization of Gaussian membership functions.
Given a set of input–output data pairs as Eq. (28), the sampling data is preprocessed, including deleting the duplicated data pairs and min–max normalization according to Eq. (29).
where (x^{(p)} in {text{R}}^{m} ;and;y^{(p)} in {text{R}}).
where (x^{prime}) is the min–max normalized set, x is a set of the observed values present in x, (x_{min }) is the minimum values in x, (x_{max }) is the maximum values in x.
The basic problem is to extract rules which describe how the output variable (y in {text{R}}) is influenced by the m input variables (x = (x_{1} , ldots ,x_{m} )^{T} in R^{m}) based on Eq. (28). FCM is a powerful unsupervised learning technique to extract rules of Mamdani structure as Eq. (1), which divides sampling data into several clusters based on their similarity5. In the proposed model, K is the number of clusters and specified to be a positive odd integer. After the sampling data is clustered by FCM algorithm, the input space of each numerical input is divided into K sections, and K Mamdani rules is generated. The rule base of Node N11 in Fig. 7 is composed by these K rules.
The semantic inputs in SET2 and outputs are Subjective Product Appraisals (SPA). For simplicity and interpretability, the input space of these semantic inputs and outputs are also divided into K sections. For example, if K = 3, the semantic choice scale is {Low, Nominal, High} presented by the positive integers 1–3. These rule bases are derived from the knowledge of experts and engineers. It is worth noting that the quality of the rule bases affects the output to some degree. The rule base for Nodes N1n and N2j (j = 1,…,n−1), which are associated with these semantic inputs, is Mamdani type as Eq. (1), and the number of rules is not over K2 because of only two inputs of each node.
The identity nodes Iij (i = 1,…,n; j = 1,…,n−1) are with the rule base of Mamdani type, too. And the number of rules is K because of only one input of each identity node.
As shown in Eq. (1), each fuzzy rule has multiple antecedents. The AND fuzzy operation intersection is implemented, which is defined as Eq. (4). The reason is that all the input variables simultaneously affect the output, and rules in antecedent must be met simultaneously in order for consequent to occur. For prediction problems, only “and” rules are required since the antecedents are different components of a single input vector.
The data base is associated with the type of membership functions. The process of assignment membership functions can be intuitive or based on some algorithmic operations. Six straightforward methods are described in the literature to assign membership functions to fuzzy variables. Based on the judgment about the probability density functions from researchers and ourselves, the shape of membership function of input variables of node N11 is Gaussian curve, as defined in Eq. (2). The position of the peak relative to the universe and the standard deviation are two parameters of a Gaussian membership function. The parameters are determined by FCM algorithm based on sampling data. In this paper, the result of FCM algorithm is named as FCM model, in which the knowledge base is with parameters of Gaussian membership functions.
For these semantic inputs, a natural membership function that readily comes to mind is symmetric triangular membership functions defined as Eq. (3). Two main reasons motivate the choice: one is their optimal interface design and the other is its semantic integrity.
In this step, the optimization is aimed towards the data base of the FCM model and based on the sampling data as Eq. (3). MATLAB-Fuzzy Logic Tool Box (genfis) is utilized to generate the FIS of FCM model. To improve the precision as well as reduce the loss of the interpretability, the optimization is only aiming towards parameters (θ and σ) of inputs and the output. The structure of this FIS is unchanged. As mentioned in section “PSO”, as a global optimization method, PSO algorithm is suitable for initial training to tune the parameters. In the proposed model, the MATLAB function “tunefis” is employed with the specified tuning algorithm name as “particleswarm”. The optimized model is named as FCM–PSO in this paper. It works by running the FCM method until it reaches its stopping criterion. Then it runs the PSO algorithm to try to achieve a better solution. In the experiments presented in section “Case study”, the following stopping criteria were used:
FCM: when there is a variation less than or equal to 1e−8 on Minimum improvement in objective function (MinImprovement for short).
PSO: when relative change in the best objective function value is less than 1e-8 on FunctionTolerance (a method option of “tunefis”) or it reaches maximum number of iterations (MaxNumIteration for short). The experiments explored 11 different MaxNumIterations in the set {10, 20, 30, 40, 50, 100, 200, 300, 400, 1000, 2000}.
The semantic inputs are organized into HFS structure as shown in Fig. 7. Two input variables (xm+1 and xm+2) are organized into N21, and other input variables are added one by one after converted by the identify node. And N2,n−1 combines all n semantic inputs. Identify nodes I1j (j = 2,…,n−1) are employed to combine N2,n−1 with N11. N1n is the final node which combine numerical inputs with semantic inputs.
To combine these FN nodes into a SFS, the rule bases of all nodes are merged vertically and horizontally39. The rule bases of N11, N21, I31, … and In,1 are merged firstly, the result named rule base V1, and in the same method, rule base V2 is from I12, N22, … and In,2, rule base Vn−1 is from the last vertical merging of I1,n−1 and N2,n−1. And then, all the rule base of Vi (0 < i ≤ n−1) are merged horizontally. Till now, the single rule base RB of the proposed model is generated, as topological expression in Eq. (30).
where “ + ” means vertical merging operation and “ ∗ ” means horizontal merging operation.
After the multiple rule bases are merged into a linguistically equivalent single rule base, the FN model is converted to a kind of SFS model with m + n inputs, a output and a knowledge base RB, as shown in Fig. 8.
SFS structure after knowledge base combination based on FN method.
In this step, the optimization is implemented to the data base of N2j (j = 1,…,n−1) in Fig. 7. The initial values of parameters (a and α) of symmetric triangular membership functions, which was specified by experts, are tuned by PSO algorithm.
To implement the tuning, SPAs of N2j should be transformed to numerical data. For example, the Eq. (31) can finish this transform. The semantic appraisal for each semantic input is from the experts’ judgments.
where i is the ith item in the linguistic term set. K is the number of items of the linguistic term set.
For example, the linguistic term set {small, average, big} have 3 items, in which the item “average” can be converted to 0.5, as shown in Fig. 9A. And {very small, small, average, big, very big} have 5 items, and the item “big” is same to 0.75, as shown in Fig. 9B. What’s more, the experts can appraise a semantic input by using a real number in the interval [0, 1] directly with the help of membership functions.
Symmetric triangular membership function with 3 linguistic terms and 5 linguistic terms respectively.
Combining ({text{D}}_{SPAi}) of xm+1 to xm+n with data pairs(x1 to xm, and y) produces new data pairs for FN. Two different type of input data are included in the new data pairs. One is from numerical inputs and this kind of inputs generally is quantifiable. For example, Lines of a paragraph of software code can be considered as numerical input. The other is from qualitative inputs. For example, programming language complexity is always appraised by semantic terms.
The new data pairs for FN is utilized to tune the parameters of triangular membership functions. The process of optimization is similar to section “Parameters optimization of Gaussian membership functions”, and with the same stopping criteria.
After being optimized in this step, the knowledge base RB of the proposed model is renewed and ready for prediction.
The proposed model with optimized knowledge base has been converted into a SFS model as shown in Fig. 8 in the previous steps. Given a data pairs of m numerical inputs and n semantic inputs, the prediction value is yielded according to the FCM-FN model.
In this section, a simple FCM-FN model was applied to 6 real world datasets to evaluate the performances. The details are described in 4 parts: the simple FCM-FN model, the performance metrics, the 6 datasets, and the results of case study. The model was implemented on a computer with an Intel Core i7–9750, 2.60 GHz processor and 16 GB RAM, running a Windows 10(64-bit) operating system and MATLAB R2019a.
Prior to building the rules, the number of linguistic variables is limited to a considerable size to avoid excessive rule explosion and deteriorated interpretation. This is because potentially, every combination of variables could require a distinct rule. Due to this, the linguistic terms used in the performance evaluation are limited to 3 which are {poor, fair and good}, or {low, moderate and high}40.
In order to show the interpretation and accuracy of the model, a simple FCM-FN model with only 3 numerical inputs and 3 semantic inputs is set up, as shown in Fig. 10. And 6 real world datasets are selected, which are descripted in section “Datasets”. For selection of the input variables x1 to x3 of each dataset, Pearson’s and Spearman’s Correlation Coefficients are utilized to select top 3 relevant input variables according to the strength and direction of a linear relationship between each input and the output. The definitions of inputs x4 to x6 depend on the needs of prediction. The data pairs of all 6 inputs and a output is nondimensionalized according to Eq. (4).
Simple FCM-FN model with 3 numerical inputs and 3 sematic inputs.
In this simple FCM-FN model, the number of linguistic sets is specified to 3(K = 3). It means that the clusters in FCM is 3, x1 to x3, z1, y are with 3 Gaussian membership functions and x4 to x6, z2, z3 are with 3 symmetric triangular membership functions, separately.
Each node of N21, N22, N13 is with 9 (= 32) rules. For example, in order to predict the software’s quality, the input variable x4 is specified as extent of the supporting document set, x5 is comprehensibility of the supporting document set and x6 is programming language complexity. They are organized into hierarchical structure, as shown in Fig. 10. The structure was not easy to make sense because z2 and z3 should be with exact connotation. Here, z2 is the subjective complexity appraisal, and z3 is subjective supporting environment appraisal. As shown in Table 4, the rule bases of N21, N22 and N13 are presented by Boolean matrices and derived by the knowledge of software engineer experts.
The knowledge base of N11 is generated by FCM algorithm, and the rule base includes 3 rules, and each identification node is also with 3 rules. Being merged based on FN method, the rule base of the simple FCM-FN model is with 81 rules.
As opposed to most existing approaches where the focus is to improve accuracy, the FCM-FN method focuses to maintain interpretation, transparency and accuracy by means of the modular rule bases that reflect the subsystems of the modeled system. There are three performance indicators to show the quality of the associated models. They are called Transparency Index (TI)22, Interpretation Index(PI) and Accuracy Index (AI).
The first performance indicator TI reflects the transparency according to the extent of its opaqueness from the inside of a model, as shown by Eq. (32).
where t is the total number of inputs, 1 is refer to the only one output, p is the number of non-identity nodes and q is the number of non-identity connections. A lower TI implies better transparency.
For example, in the simple FCM-FN model as shown in Fig. 10, t = 6, p = 4, q = 3,and TI = 1.
The second performance indicator PI reflects the interpretability of a model. It’s obvious that transparency is helpful to interpretability and more parameters impact the interpretability. Thus, we put forward a modified TI to reflects the interpretability of a FCM-FN model, as shown by Eq. (33).
where r is the total number of parameters, s is the total number of optimized parameters. A lower PI implies better interpretability.
For example, in the model showed in Fig. 10, t = 6, p = 4, q = 3, r = 20, s = 14, and PI = 1.646.
The third performance indicator AI reflects the accuracy of a model by means of Symmetric Mean Absolute Percentage Error (SMAPE) as shown by Eq. (35) and Goodness-of-Fit (R2) as shown by Eq. (9).
where n is the total number of observations compared, actual is the ith value of the test dataset estimated values vector, and prediction is the ith value of output vector of a model, mean is the mean value of actual. Lower SMAPE or higher R2 implies better accuracy.
Six well-known real-world datasets are loaded to evaluate the performances of the FCM-FN model compared with FCM model and FCM-PSO model (See Supplementary File – raw dataset).
Table 5 is a summary of the 6 datasets. These datasets are cases proposed in previous studies in various fields including social, material, medical and software. Based on Pearson’s and Spearman’s Correlation Coefficients, only top 3 attributes and the prediction target of each database are selected as simplified benchmark dataset to evaluate the performance. The number of instances is the rows of original databases, and the number of unique instances is the rows of benchmark dataset after deleted the duplicated instances.
White wine dataset and red wine dataset were obtained from the same source and with the same structure41. The two datasets were related to white and red variants of the Portuguese “Vinho Verde” wine. Both datasets contain chemical analysis of wine derived from three different cultivars grown in the same region. Eleven physicochemical attributes of wine were included: 1—fixed acidity, 2—volatile acidity, 3—citric acid, 4—residual sugar, 5—chlorides, 6—free sulfur dioxide, 7—total sulfur dioxide, 8—density, 9—pH, 10—sulphates, 11—alcohol. The prediction target was wine quality, which was graded between 0 (very bad) and 10 (very excellent) by wine experts.
Concrete dataset recorded 1030 instances and 9 attributes. The prediction target was concrete compressive strength, and the 8 input variables including 1—cement, 2—blast furnace slag, 3—fly ash, 4—water, 5—super plasticizer, 6—coarse aggregate, 7—fine aggregate and 8-age. The concrete compressive strength was a highly nonlinear function of age and 7 ingredients. And the actual concrete compressive strength (MPa) for a given mixture under a specific age (days) was determined from laboratory42.
Boston dataset recorded 506 house prices. The dataset was taken from the StatLib library which was maintained at Carnegie Mellon University. The original dataset contained 14 attributes. In order to avoid racism, only 13 attributes was contained in the newly version dataset: 1—CRIM (per capita crime rate by town), 2—ZN (proportion of residential land zoned for lots over 25,000 sq.ft.), 3—INDUS (proportion of non-retail business acres per town), 4—CHAS (Charles River dummy variable), 5—NOX(nitric oxides concentration), 6—RM (average number of rooms per dwelling), 7—AGE (proportion of owner-occupied units built prior to 1940), 8—DIS (weighted distances to five Boston employment centres), 9—RAD (index of accessibility to radial highways), 10—TAX (full-value property-tax rate per $10,000), 11—PTRATIO (pupil-teacher ratio by town), 12—LSTAT (lower status of the population), 13—MEDV (Median value of owner-occupied homes in $1000’s). And MEDV was usually the prediction target.
Diabetes recorded 442 patients health conditions by 11 attributes: 1—age, 2—sex, 3—bmi (Body mass index), 4—bp (Average Blood Pressure), 5—s1, 6—s2, 7—s3, 8—s4, 9—s5, 10—s6, 11—outcome. And s1–s6 were 6 serum test data one year later. The prediction target was outcome (quantitative indicator of diabetes one year later).
QUES database was selected from popular object-oriented maintainability datasets which were published by Li and Henry43. Ten attributes were included as input variables: 1—WMC (Weighted method per class), 2—DIT (Depth of the inheritance tree), 3—NOC (Number of children), 4—RFC (Response for class), 5—LCOM (Lack of cohesion of methods), 6—MPC (Message-passing coupling), 7—DAC (Data abstraction coupling), 8—NOM (Number of methods), 9—SIZE2 (Number of properties), 10—SIZE1 (Lines of code). The prediction target was CHANGE (Number of lines changed in the class), which recorded the number of changed lines in the code during a three-year maintenance period. The values of 11 attributes in QUES was collected from 71 classes in a software system.
In this section, the results of the proposed model on the real-world datasets are presented. The comparison is implemented among FCM model, FCM-PSO model and FCM-FN model. FCM model is only related to the node N11. After the parameters being tuned, FCM model is transformed into FCM-PSO. And after implementing the FN method and PSO tuning, FCM-PSO model is transformed into FCM-FN.
Table 6 summarized the performance TI and PI of the three models: FCM, FCM-PSO, FCM-FN. TI is related to the structure of models directly. The FCM and FCM-PSO have the same structure and TI (= 4). By contrast, the FCM-FN model is with non-identity nodes and connections and lower TI (= 1). PI is related to both the structure and parameters. Though FCM and FCM-PSO have the same structure, but the latter tunes the parameters and then impact PI, which is added from 5.021 to 5.660. Due to lower TI, the PI of FCM-FN is 1.646 and far lower than the other two models.
Table 7 are the results of performance SMAPE of the three models utilizing the 6 benchmark datasets. The AI of each model is appraised with the process of ten-fold cross-validation. Each dataset is randomly shuffled and then divided into 10 partitions. 9 partitions for training and 1 partition for testing. The value of AI is the mean value after running each model 10 times, in format of percentage. In Table 7, the ‘%’ is omitted in order to save space, and the values of best accuracy among the three models are bolded.
In order to explore the effects of different iterations of PSO, 11 different values from 10 to 2000 are set separately as the MaxNumIteration parameter of PSO. The values of SMAPE after 11 different iterations are shown in Table 7. It’s obvious that the accuracy of FN-PSO and FCM-PSO is far better than FCM model, and FN-PSO better than FCM-PSO in most instances.
The means of 11 SMAPE values of each model is shown in Table 8, as well as the improvements of FCM-FN compared with FCM and FCM-PSO respectively. The accuracy of the FCM-FN model is improved from 5.012% to 30.784% compared with FCM, and from 0.204% to 1.201% compared with FCM-PSO. The average SMAPE of each dataset shows that FCM-FN models marginally outperforms FCM-PSO models. This observation is not surprising since FCM-FN model derived most of its initial semantic inputs from the experts’ knowledge.
In general, increasing number of iterations in parameter-tuning phase, the SMAPE and R2 is increasing accordingly as shown in Fig. 11A–C and Table 7. But, as shown in Fig. 11D–F, the R2 is not always increasing along with the iterations. What’s more, as shown in Fig. 11F, the R2 fluctuates sharply because there are only 71 instances. As shown in Table 7, the PSO algorithm always converged before reaching the aforementioned MaxNumIteration. For example, The FN-PSO and FCM-PSO reached their best accuracy when the benchmark dataset was red wine and MaxNumIteration was 200, and more MaxNumIteration was not yielded higher accuracy because the FCM-PSO converged no more than 257 iterations and FN-PSO no more than 51 iterations. And doing so increased the duration of the optimization process and also increased validation error due to overtuned system parameters with the training data. Thus, the following accuracy values fluctuated around the best one.
The R2 of the three models after different iterations’ optimization.
To improve the performances of prediction model both in accuracy and transparency with well tradeoff interpretability, this paper proposed a FCM-FN model which combines FCM, FN and PSO methods. The main motivation of utilizing FN in prediction model is that interpretability is the dominant feature of a fuzzy model in security-oriented field and other special fields. As noted earlier, FN method has been characterized by the nature of a white-box where the output variable map the input metrics by connections, and the implied information in the FN model has been shown totally. As to the performance of TI and PI, FCM-FN model is overwhelming compared to FCM model or FCM-PSO model.
Based on FCM method, the numerical data is clustered and then denoted by rules with simple and interpretable structure. Optimized with PSO method, the accuracy of FCM-FN model has been improved significantly (over 5%).
In conclusion, the FCM-FN model has more transparency, interpretability and accuracy than FCM model or FCM-PSO model. Moreover, FCM-FN model is a special type of SFS, because the numerical data and linguistic appraisals can be combined into the prediction model at the same time. Although this strategy is not perfect, it tends to catch more subjective appraisals which have shown their powerful efficacy in previous researches44,45.
In this paper only 3 subjective appraisal variables are specified as the example, and the results indicate that it is possible to find qualitative rules. We do not claim that the rules between the 3 variables are the most suitable ones, nor that the FCM-FN based models overwhelm the earlier models in terms of performances other than interpretability and accuracy (SMAPE and R2). Our results showed that FN based models improves the interpretability and accuracy significantly. Notwithstanding its limitation, it is possible to improve the accuracy by implementing more linguistic terms and more accurate subjective appraisals in FCM-FN models. Despite its preliminary character, FCM-FN model is effective for prediction.
To optimize this model in the future, we would like to conduct experiment from the following four aspects: First, integrate type-2 fuzzy system and put it forward in FCM-FN models. Second, integrate other optimization methods, such as GA and “Patternsearch”. As an important optimization method in function “tunefis”, “Patternsearch” performs better for small parameter ranges since they are local optimizers. “Patternsearch” may produce faster convergence compared to particleswarm as the tradeoff for accuracy. Third, change the number of linguistic terms or the number of clusters in FCM to explore their influence on accuracy. Forth, integrate newly fuzzy clustering methods to explore their influence on accuracy and interpretability.
The datasets used and analyzed during the current study are available from three different ways. Three datasets (white wine, red wine and concrete) are available from the UCI Machine Learning Repository (http://archive.ics.uci.edu/ml/index.php), the datasets (boston and diabetes) are available from kaggle website (https://www.kaggle.com/datasets/) and the QUES dataset originally published by Li and Henry43.
The MATLAB source code for this project is available upon reasonable request. Please contact Xiaowei Wang ([email protected]).
Wang, G., Wang, H. & Long, Z. Norm approximation of mamdani fuzzy system to a class of integrable functions. Int. J. Fuzzy Syst. 23, 833–848. https://doi.org/10.1007/s40815-020-01008-3 (2021).
Article Google Scholar
Tsekouras, G., Sarimveis, H., Kavakli, E. & Bafas, G. A hierarchical fuzzy-clustering approach to fuzzy modeling. Fuzzy Sets Syst. 150, 245–266. https://doi.org/10.1016/j.fss.2004.04.013 (2005).
MathSciNet Article MATH Google Scholar
Zhang, Y., Wu, X. B., Xing, Z. Y. & Hu, W. L. On generating interpretable and precise fuzzy systems based on Pareto multi-objective cooperative co-evolutionary algorithm. Appl. Soft Comput. 11, 1284–1294. https://doi.org/10.1016/j.asoc.2010.03.005 (2011).
Article Google Scholar
Zador, A. M. A critique of pure learning and what artificial neural networks can learn from animal brains. Nat. Commun. 10, 1–7. https://doi.org/10.1038/s41467-019-11786-6 (2019).
ADS CAS Article Google Scholar
Dick, S., Meeks, A., Last, M., Bunke, H. & Kandel, A. Data mining in software metrics databases. Fuzzy Sets Syst. 145, 81–110. https://doi.org/10.1016/j.fss.2003.10.006 (2004).
MathSciNet Article Google Scholar
Guillaume, S. Designing fuzzy inference systems from data: An interpretability-oriented review. Fuzzy Syst. IEEE Trans. 9, 426–443. https://doi.org/10.1109/91.928739 (2001).
Article Google Scholar
Jiang, G. C. & Sun, Z. J. Improvement on a fuzzy C-means algorithm based on genetic algorithm. Appl. Mech. Mater. 614, 385–388. https://doi.org/10.4028/www.scientific.net/AMM.614.385 (2014).
Article Google Scholar
Ding, Y. & Fu, X. Kernel-based fuzzy c-means clustering algorithm based on genetic algorithm. Neurocomputing 188, 233–238. https://doi.org/10.1016/j.neucom.2015.01.106 (2016).
Article Google Scholar
Silva-Filho, T. M., Pimentel, B. A., Souza, R. M. C. R. & Oliveira, A. L. I. Hybrid methods for fuzzy clustering based on fuzzy c-means and improved particle swarm optimization. Expert Syst. Appl. 42, 6315–6328. https://doi.org/10.1016/j.eswa.2015.04.032 (2015).
Article Google Scholar
Kuo, R. J., Lin, T. C., Zulvia, F. E. & Tsai, C. Y. A hybrid metaheuristic and kernel intuitionistic fuzzy c-means algorithm for cluster analysis. Appl. Soft Comput. 67, 299–308. https://doi.org/10.1016/j.asoc.2018.02.039 (2018).
Article Google Scholar
Izakian, H. & Abraham, A. Fuzzy C-means and fuzzy swarm for fuzzy clustering problem. Expert Syst. Appl. 38, 1835–1838. https://doi.org/10.1016/j.eswa.2010.07.112 (2011).
Article Google Scholar
Hu, J. & Yin, H. FCM clustering algorithm based on PSO-TVAC algorithm with adaptively weighted centers. Adv. Appl. Math. 10, 953–962. https://doi.org/10.12677/AAM.2021.104104 (2021).
Article Google Scholar
Zhang, Y. C., Xiong, X. & Zhang, Q. D. An improved self-adaptive PSO algorithm with detection function for multimodal function optimization problems. Math. Probl. Eng. 657–675, 2013. https://doi.org/10.1155/2013/716952 (2013).
MathSciNet Article MATH Google Scholar
Alam, S., Dobbie, G., Koh, Y. S., Riddle, P. & Rehman, S. U. Research on particle swarm optimization based clustering: A systematic review of literature and techniques. Swarm Evol. Comput. 17, 1–13. https://doi.org/10.1016/j.swevo.2014.02.001 (2014).
Article Google Scholar
Chen, S., Xu, Z. & Tang, Y. A hybrid clustering algorithm based on fuzzy c-means and improved particle swarm optimization. Arab. J. Sci. Eng. 39, 8875–8887. https://doi.org/10.1007/s13369-014-1424-9 (2014).
MathSciNet Article MATH Google Scholar
Ahmed, M. A. & Al-Jamimi, H. A. Machine learning approaches for predicting software maintainability: A fuzzy-based transparent model. IET Softw. 7, 317–326. https://doi.org/10.1049/iet-sen.2013.0046 (2013).
Article Google Scholar
Wang, L.-X. & Mendel, J. M. Generating fuzzy rules by learning from examples. IEEE Trans. Syst. Man Cybern. 22, 1414–1427. https://doi.org/10.1109/21.199466 (1992).
MathSciNet Article Google Scholar
Wang, X., Gegov, A., Farzad, A., Chen, Y. & Hu, Q. Fuzzy network based framework for software maintainability prediction. Internat. J. Uncertain. Fuzziness Knowl.-Based Syst. 27, 841–862. https://doi.org/10.1142/s0218488519500375 (2019).
CAS Article Google Scholar
Bucolo, M., Fortuna, L. & Rosa, M. L. Complex dynamics through fuzzy chains. IEEE Trans. Fuzzy Syst. 12, 289–295. https://doi.org/10.1109/TFUZZ.2004.825969 (2004).
Article Google Scholar
Lee, M.-L., Chung, H.-Y. & Yu, F.-M. Modeling of hierarchical fuzzy systems. Fuzzy Sets Syst. 138, 343–361. https://doi.org/10.1016/S0165-0114(02)00517-1 (2003).
MathSciNet Article Google Scholar
Joo, M. G. & Lee, J. S. A class of hierarchical fuzzy systems with constraints on the fuzzy rules. IEEE Trans. Fuzzy Syst. 13, 194–203. https://doi.org/10.1109/TFUZZ.2004.840096 (2005).
Article Google Scholar
Gegov, A., Arabikhan, F. & Petrov, N. Linguistic composition based modelling by fuzzy networks with modular rule bases. Fuzzy Sets Syst. 269, 1–29. https://doi.org/10.1016/j.fss.2014.06.014 (2014).
MathSciNet Article MATH Google Scholar
Yaakob, A. M., Gegov, A. & Rahman, S. Selection of alternatives using fuzzy networks with rule base aggregation. Fuzzy Sets Syst. 341, 123–144. https://doi.org/10.1016/j.fss.2017.05.027 (2018).
MathSciNet Article MATH Google Scholar
Gegov, A., Sanders, D. & Vatchova, B. Mamdani fuzzy networks with feedforward rule bases for complex systems modelling. J. Intell. Fuzzy Syst. 30, 2623–2637. https://doi.org/10.3233/IFS-151911 (2016).
Article Google Scholar
Gegov, A. Fuzzy Networks for Complex Systems: A Modular Rule Base Approach. https://doi.org/10.1007/978-3-642-15600-7 (Springer, 2010).
Book MATH Google Scholar
Mendel, J. M. Uncertain Rule-Based Fuzzy Logic Systems: Introduction and New Directions 2nd edn. (Springer, 2017).
Book Google Scholar
Zadeh, L. A. Outline of a new approach to the analysis of complex systems and decision processes. IEEE Trans. Syst. Man Cybern. 3, 28–44. https://doi.org/10.1109/TSMC.1973.5408575 (1973).
MathSciNet Article MATH Google Scholar
Zadeh, L. A. Fuzzy sets. Inf. Control 8, 338–353. https://doi.org/10.1016/S0019-9958(65)90241-X (1965).
Article MATH Google Scholar
Mamdani, E. H. Application of fuzzy logic to approximate reasoning using linguistic synthesis. Comput. IEEE Trans. 100, 1182–1191. https://doi.org/10.1109/TC.1977.1674779 (1977).
Article MATH Google Scholar
Full, J. C. B. R. E. W. FCM: The fuzzy c -means clustering algorithm. Comput. Geosci. 10, 191–203. https://doi.org/10.1016/0098-3004(84)90020-7 (1984).
ADS Article Google Scholar
Askari, S., Montazerin, N. & Zarandi, M. F. Generalized possibilistic fuzzy c-means with novel cluster validity indices for clustering noisy data. Appl. Soft Comput. 53, 262–283. https://doi.org/10.1016/j.asoc.2016.12.049 (2017).
Article Google Scholar
Rubio, E. et al. An extension of the fuzzy possibilistic clustering algorithm using type-2 fuzzy logic techniques. Adv. Fuzzy Syst. 2017, 53. https://doi.org/10.1155/2017/7094046 (2017).
Article Google Scholar
Mansoureh, N., Mohammad HosseinFazel, Z. & Susan, B. A fuzzy cluster-validity index based on the topology structure and node attribute in complex networks. Expert Syst. Appl. 187, 115913. https://doi.org/10.1016/j.eswa.2021.115913 (2022).
Article Google Scholar
Kaixin, Z., Yaping, D., Zhiyang, J. & Ye, J. General fuzzy C-means clustering algorithm using Minkowski metric. Signal Process. 188, 108161. https://doi.org/10.1016/j.sigpro.2021.108161 (2021).
Article Google Scholar
Filev, D. Fuzzy modeling of complex systems. Int. J. Approx. Reason. 5, 281–290. https://doi.org/10.1016/B978-1-4832-1450-4.50046-8 (1991).
Article MATH Google Scholar
Fernández Caballero, A. Contribution of fuzziness and uncertainty to modern artificial intelligence. Fuzzy Sets Syst. 160, 129. https://doi.org/10.1016/j.fss.2008.05.002 (2009).
MathSciNet Article Google Scholar
Aja-Fernandez, S. & Alberola-Lopez, C. Matrix modeling of hierarchical fuzzy systems. IEEE Trans. Fuzzy Syst. 16, 585–599. https://doi.org/10.1109/TFUZZ.2007.905905 (2008).
Article Google Scholar
Hall, L. O. Rule chaining in fuzzy expert systems. IEEE Trans. Fuzzy Syst. 9, 822–828. https://doi.org/10.1109/91.971731 (2001).
Article Google Scholar
Gegov, A., Petrov, N. & Vatchova, B. In Intelligent Systems (IS), 2010 5th IEEE International Conference 197–202 (IEEE, 2010).
Olugu, E. U. & Wong, K. Y. An expert fuzzy rule-based system for closed-loop supply chain performance assessment in the automotive industry. Expert Syst. Appl. 39, 375–384. https://doi.org/10.1016/j.eswa.2011.07.026 (2012).
Article Google Scholar
Cortez, P., Cerdeira, A., Almeida, F., Matos, T. & Reis, J. Modeling wine preferences by data mining from physicochemical properties. Decis. Support Syst. 47, 547–553. https://doi.org/10.1016/j.dss.2009.05.016 (2009).
Article Google Scholar
Yeh, I.-C. Modeling of strength of high-performance concrete using artificial neural networks. Cem. Concr. Res. 28, 1797–1808. https://doi.org/10.1016/s0008-8846(98)00165-3 (1998).
CAS Article Google Scholar
Li, W. & Henry, S. Object-oriented metrics that predict maintainability. J. Syst. Softw. 23, 111–122. https://doi.org/10.1016/0164-1212(93)90077-B (1993).
Article Google Scholar
Van den Broeke, M., De Baets, S., Vereecke, A., Baecke, P. & Vanderheyden, K. Judgmental forecast adjustments over different time horizons. Omega 87, 34–45. https://doi.org/10.1016/j.omega.2018.09.008 (2019).
Article Google Scholar
Arvan, M., Fahimnia, B., Reisi, M. & Siemsen, E. Integrating human judgement into quantitative forecasting methods: A review. Omega 86, 237–252. https://doi.org/10.1016/j.omega.2018.07.012 (2019).
Article Google Scholar
Download references
The first author would like to thank Academic Leader Program (grant no. JXSGZL202203) for their financial support, thank Dr. Ting Yang and Dr. Qing Liu’s advices, and thank Dr. Alexander Gegov and the School of Computing at University of Portsmouth who offered the invitation as a visiting researcher and shared the theory of Fuzzy Network. Thank two anonymous reviewers for the constructive comments and valuable advices.
College of Power Engineering, Naval University of Engineering, Wuhan, 430033, China
Xiaowei Wang, Yanqiao Chen & Jiashan Jin
Army Engineering University of PLA, Wuhan, 430075, China
Xiaowei Wang & Baohua Zhang
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
X.W. and Y.C. wrote the main manuscript text, and J.J. and B.Z. implemented the case study. All authors reviewed the manuscript.
Correspondence to Yanqiao Chen.
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
Wang, X., Chen, Y., Jin, J. et al. Fuzzy-clustering and fuzzy network based interpretable fuzzy model for prediction. Sci Rep 12, 16279 (2022). https://doi.org/10.1038/s41598-022-20015-y
Download citation
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-022-20015-y
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
© 2022 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.
How Mercedes is using software to rewrite product development playbook – Automotive News Europe
Christoph Starzynski, who leads the development of Mercedes’ EQ subbrand, says the automaker is investing in startups “to see what’s the best horse in the race.”
DENVER — More than a decade ago, venture capitalist Marc Andreessen famously claimed, “Software is eating the world.” That trend certainly is reverberating through the automotive world.
What was once a primarily atoms-focused hardware business is increasingly morphing into a bits-driven software and services one. Powertrain electrification and in-vehicle connectivity have turned cars into computers on wheels, with the average vehicle running more than 100 million lines of software code.
Christoph Starzynski, who leads the development of Mercedes-Benz‘s electric EQ subbrand, has a front-seat view of the industry’s tech transformation as he helps navigate the German luxe brand toward an all-EV lineup.
In a software-first world, automakers can deliver new functionality and dial up performance via over-the-air updates long after the vehicle has rolled off the assembly line, Starzynski told Automotive News at a press event here.
“How can we, with incremental steps, improve the product without asking the customer to come into the workshop?” he said.
In a world of “software-defined vehicles,” old-school automakers are competing with new-school Silicon Valley for computer engineers, user-experience developers and coders.
General Motors said it would hire more than 8,000 tech workers this year to accelerate the development of electric vehicle and software-driven services. Mercedes, meanwhile, is recruiting an army of 10,000 software engineers in Berlin, China, India, Israel, Japan and the United States.
In a survey of 572 auto executives by research institute Capgemini, 97 percent said that four out of 10 in-house workers would need software skills within five years, from IT architects to cloud management professionals to cybersecurity experts.
Automakers are rewriting their product development playbook to prioritize speed and collaboration.
“We’re in a different competition right now,” Starzynski said. “We looked around and asked, ‘Whom can we learn from internally?’ ”
Mercedes found inspiration in its Formula 1 racing team, which relies on “continuous iteration” to shave milliseconds off lap times each weekend.
For EQ model line development, Mercedes has ditched department silos in favor of a flat organizational structure that encourages real-time collaboration.
“We brought people together from all areas, and they were fighting for the best ideas,” Starzynski. “Then we pulled management in” at the board level for quicker decision making.
The new workflow manifests in Mercedes’ development hubs from Shanghai to Sunnyvale, California. Open-floor workspaces foster serendipitous interactions and spark idea-sharing between teams.
Mercedes also looks beyond its R&D labs to stay ahead in the auto industry’s technology arms race.
Starzynski said the automaker is investing in startups “to see what’s the best horse in the race.”
Last year, Mercedes acquired British startup Yasa. The Oxford University spinoff has developed a high-torque, lightweight e-motor that delivers more power with greater efficiency.
Yasa’s axial flux motor technology will power Mercedes’ AMG electric-only platform and other future models.
Meanwhile, an investment in Alameda, Calif.-based Sila could bring highly energy-dense battery technology to an electric G-Class wagon around mid-decade.
The Sila battery uses silicon-based anodes and is 20 to 40 percent more energy dense than comparable cells on the market, Mercedes said.
Starzynski said the boxy, off-roader will be the “perfect car” for the technology.
Please enter a valid email address.
Please enter your email address.
Please verify captcha.
Please select at least one newsletter to subscribe.
You can unsubscribe at any time through links in these emails. For more information, see our Privacy Policy.
Sign up and get the best of Automotive News Europe delivered straight to your email inbox, free of charge. Choose your news – we will deliver.
You can unsubscribe at any time through links in these emails. For more information, see our Privacy Policy.
Get 24/7 access to in-depth, authoritative coverage of the auto industry from a global team of reporters and editors covering the news that’s vital to your business.
Founded in 1996, Automotive News Europe is the preferred information source for decision-makers and opinion leaders operating in Europe.
1155 Gratiot Avenue
Detroit MI 48207-2997
Tel: +1 877-812-1584
Email Us
ISSN 2643-6590 (print)
ISSN 2643-6604 (online)