PRESS-NEWS.org - Press Release Distribution
PRESS RELEASES DISTRIBUTION

To excel at engineering design, generative AI must learn to innovate, study finds

AI models that prioritize similarity falter when asked to design something completely new.

To excel at engineering design, generative AI must learn to innovate, study finds
2023-10-19
(Press-News.org)

ChatGPT and other deep generative models are proving to be uncanny mimics. These AI supermodels can churn out poems, finish symphonies, and create new videos and images by automatically learning from millions of examples of previous works. These enormously powerful and versatile tools excel at generating new content that resembles everything they’ve seen before. 

But as MIT engineers say in a new study, similarity isn’t enough if you want to truly innovate in engineering tasks. 

“Deep generative models (DGMs) are very promising, but also inherently flawed,” says study author Lyle Regenwetter, a mechanical engineering graduate student at MIT. “The objective of these models is to mimic a dataset. But as engineers and designers, we often don’t want to create a design that’s already out there.”

He and his colleagues make the case that if mechanical engineers want help from AI to generate novel ideas and designs, they will have to first refocus those models beyond “statistical similarity.” 

“The performance of a lot of these models is explicitly tied to how statistically similar a generated sample is to what the model has already seen,” says co-author Faez Ahmed, assistant professor of mechanical engineering at MIT. “But in design, being different could be important if you want to innovate.”

In their study, Ahmed and Regenwetter reveal the pitfalls of deep generative models when they are tasked with solving engineering design problems. In a case study of bicycle frame design, the team shows that these models end up generating new frames that mimic previous designs but falter on engineering performance and requirements. 

When the researchers presented the same bicycle frame problem to DGMs that they specifically designed with engineering-focused objectives, rather than only statistical similarity, these models produced more innovative, higher-performing frames. 

The team’s results show that similarity-focused AI models don’t quite translate when applied to engineering problems. But, as the researchers also highlight in their study, with some careful planning of task-appropriate metrics, AI models could be an effective design “co-pilot.” 

“This is about how AI can help engineers be better and faster at creating innovative products,” Ahmed says. “To do that, we have to first understand the requirements. This is one step in that direction.”

The team’s new study appeared recently online, and will be in the December print edition of the journal Computer Aided Design. The research is a collaboration between computer scientists at MIT-IBM Watson AI Lab and mechanical engineers in MIT’s DeCoDe Lab. The study’s co-authors include Akash Srivastava and Dan Gutreund at the MIT-IBM Watson AI Lab.

Framing a problem

As Ahmed and Regenwetter write, DGMs are “powerful learners, boasting unparalleled ability” to process huge amounts of data. DGM is a broad term for any machine-learning model that is trained to learn distribution of data and then use that to generate new, statistically similar content. The enormously popular ChatGPT is one type of deep generative model known as a large language model, or LLM, which incorporates natural language processing capabilities into the model to enable the app to generate realistic imagery and speech in response to conversational queries. Other popular models for image generation include DALL-E and Stable Diffusion.

Because of their ability to learn from data and generate realistic samples, DGMs have been increasingly applied in multiple engineering domains. Designers have used deep generative models to draft new aircraft frames, metamaterial designs, and optimal geometries for bridges and cars. But for the most part, the models have mimicked existing designs, without improving the performance on existing designs.

“Designers who are working with DGMs are sort of missing this cherry on top, which is adjusting the model’s training objective to focus on the design requirements,” Regenwetter says. “So, people end up generating designs that are very similar to the dataset.”

In the new study, he outlines the main pitfalls in applying DGMs to engineering tasks, and shows that the fundamental objective of standard DGMs does not take into account specific design requirements. To illustrate this, the team invokes a simple case of bicycle frame design and demonstrates that problems can crop up as early as the initial learning phase. As a model learns from thousands of existing bike frames of various sizes and shapes, it might consider two frames of similar dimensions to have similar performance, when in fact a small disconnect in one frame — too small to register as a significant difference in statistical similarity metrics — makes the frame much weaker than the other, visually similar frame. 

Beyond “vanilla”

The researchers carried the bicycle example forward to see what designs a DGM would actually generate after having learned from existing designs. They first tested a conventional “vanilla” generative adversarial network, or GAN — a model that has widely been used in image and text synthesis, and is tuned simply to generate statistically similar content. They trained the model on a dataset of thousands of bicycle frames, including commercially manufactured designs and less conventional, one-off frames designed by hobbyists. 

Once the model learned from the data, the researchers asked it to generate hundreds of new bike frames. The model produced realistic designs that resembled existing frames. But none of the designs showed significant improvement in performance, and some were even a bit inferior, with heavier, less structurally sound frames. 

The team then carried out the same test with two other DGMs that were specifically designed for engineering tasks. The first model is one that Ahmed previously developed to generate high-performing airfoil designs. He built this model to prioritize statistical similarity as well as functional performance. When applied to the bike frame task, this model generated realistic designs that also were lighter and stronger than existing designs. But it also produced physically “invalid” frames, with components that didn’t quite fit or overlapped in physically impossible ways. 

“We saw designs that were significantly better than the dataset, but also designs that were geometrically incompatible because the model wasn’t focused on meeting design constraints,” Regenwetter says. 

The last model the team tested was one that Regenwetter built to generate new geometric structures. This model was designed with the same priorities as the previous models, with the added ingredient of design constraints, and prioritizing physically viable frames, for instance, with no disconnections or overlapping bars. This last model produced the highest-performing designs, that were also physically feasible. 

“We found that when a model goes beyond statistical similarity, it can come up with designs that are better than the ones that are already out there,” Ahmed says. “It’s a proof of what AI can do, if it is explicitly trained on a design task.”

For instance, if DGMs can be built with other priorities, such as performance, design constraints, and novelty, Ahmed foresees “numerous engineering fields, such as molecular design and civil infrastructure, would greatly benefit. By shedding light on the potential pitfalls of relying solely on statistical similarity, we hope to inspire new pathways and strategies in generative AI applications outside multimedia.”

###

Written by Jennifer Chu, MIT News

Paper: “Beyond Statistical Similarity: Rethinking Metrics for Deep Generative Models in Engineering Design”

https://www.sciencedirect.com/science/article/abs/pii/S0010448523001410

END


[Attachments] See images for this press release:
To excel at engineering design, generative AI must learn to innovate, study finds

ELSE PRESS RELEASES FROM THIS DATE:

Startup workers flee for bigger, more established companies during pandemic

Startup workers flee for bigger, more established companies during pandemic
2023-10-19
October 19, 2023 Startup workers flee for bigger, more established companies during pandemic Findings reveal vulnerability of early-stage firms in downturns Toronto - The world may have felt like it had stopped in the pandemic’s first weeks. But a “flight to safety” was underway at a popular digital job platform catering to the startup sector. Digging into the data for nearly 180,000 users from AngelList Talent (now called Wellfound), the biggest online recruitment platform for private and entrepreneurial companies, researchers have found that U.S. job hunters turned away from smaller, early-stage companies in favour of positions at bigger, more established firms. Just ...

Research repository arXiv receives $10M for upgrades

2023-10-19
ITHACA, N.Y. -- Cornell Tech has announced a total of more than $10 million in gifts and grants from the Simons Foundation and the National Science Foundation, respectively, to support arXiv, a free distribution service and open-access archive for scholarly articles. The funding will allow the growing repository with more than 2 million articles to migrate to the cloud and modernize its code to ensure reliability, fault tolerance and accessibility for researchers. “I am deeply grateful for this tremendous support from both the Simons Foundation ...

Electrons are quick-change artists in molten salts, chemists show

Electrons are quick-change artists in molten salts, chemists show
2023-10-19
In a finding that helps elucidate how molten salts in advanced nuclear reactors might behave, scientists have shown how electrons interacting with the ions of the molten salt can form three states with different properties. Understanding these states can help predict the impact of radiation on the performance of salt-fueled reactors. The researchers, from the Department of Energy’s Oak Ridge National Laboratory and the University of Iowa, computationally simulated the introduction of an excess electron into molten zinc chloride salt to see what would happen. They found three possible scenarios. In one, the electron becomes part of a molecular radical that ...

Communities of color experienced fear and mistrust of institutions during COVID-19 pandemic

Communities of color experienced fear and mistrust of institutions during COVID-19 pandemic
2023-10-19
RIVERSIDE, Calif. -- A study led by researchers in the School of Medicine at the University of California, Riverside, has found that in communities of color in Inland Southern California, historical, cultural, and social traumas induce fear and mistrust in public health and medical, scientific, and governmental institutions, which, in turn, influence these communities’ hesitation to get tested and vaccinated for COVID-19.  The study, published in Culture, Medicine, and Psychiatry, underscores the need for community-based health interventions that consider structural and social determinants of health ...

Nail salon and other small beauty service workers face significant daily health challenges

2023-10-19
The beauty service microbusiness industry in the United States — such as the small, independently-owned nail salons found across the country — is huge, with more than $62 billion in annual sales. However, most of the workers who provide these highly sought services are Asian female immigrants who earn very low wages. These workers face numerous workplace health challenges stemming from the chemicals they use, repetitive movements with handheld tools and awkward body posturing. They also are reluctant to bring attention to these conditions due to factors such as possible immigration-related trauma, lack of English proficiency, ...

Pandemic prevention consortium announces new leadership team

Pandemic prevention consortium announces new leadership team
2023-10-19
Recognizing the many milestones it has reached in recent months, Strategies to Prevent Spillover, or STOP Spillover, a project funded by the U.S. Agency for International Development (USAID) and led by Tufts University, has announced that the interim leadership team that was put in place in March 2023 will take on a permanent role for the next two years of the project. Hellen Amuguni, an associate professor in the Department of Infectious Disease and Global Health at Cummings School of Veterinary Medicine, is the new project director. The co-deputy directors are Felicia Nutter, director of the International Veterinary ...

mRNA delivered by extracellular vesicles induces immunotherapy response in glioblastoma

mRNA delivered by extracellular vesicles induces immunotherapy response in glioblastoma
2023-10-19
HOUSTON ― A team of researchers at The University of Texas MD Anderson Cancer Center has developed a new method for using extracellular vesicles to enhance responses to immunotherapy in glioblastoma, potentially opening the door for wider use of engineered messenger RNA (mRNA) for cancer therapy. The study was published today in Nature Communications. Earlier this year, a team of researchers led by Betty Kim, M.D., Ph.D., and Wen Jiang, M.D., Ph.D.,developed a novel method for loading mRNA into extracellular vesicles, small structures created by cells to transport biomolecules and nucleic acids within ...

Point-of-care technology initiative awarded $8.9 million renewal

Point-of-care technology initiative awarded $8.9 million renewal
2023-10-19
UMass Chan and UMass Lowell’s point-of-care technology initiative awarded $8.9 million renewal UMass Chan Medical School and UMass Lowell have received an $8.9 million award from the National Institutes of Health (NIH) for renewed support of their initiative to advance the development of home-based and point-of-care health technologies. The program aims to jumpstart new tools to address heart, lung, blood and sleep disorders, especially in underserved populations. The Center for Advancing Point of Care Technologies ...

GLS2 shapes ferroptosis in hepatocellular carcinoma

GLS2 shapes ferroptosis in hepatocellular carcinoma
2023-10-19
“[...] we hope that our findings will inform future decisions regarding treatment of liver disease.” BUFFALO, NY- October 19, 2023 – A new editorial paper was published in Oncotarget's Volume 14 on October 19, 2023, entitled, “GLS2 shapes ferroptosis in hepatocellular carcinoma.” In their new editorial, researchers Sawako Suzuki, Divya Venkatesh, Tomoaki Tanaka, and Carol Prives from Columbia University discuss ferroptosis regulation of GLS2 as a potential therapeutic strategy against liver diseases. “More than a decade has passed since our group (1) as well as ...

Groundbreaking study on bilingual children with Developmental Language Disorder

Groundbreaking study on bilingual children with Developmental Language Disorder
2023-10-19
Amanda Owen Van Horne sits on the floor while a child mixes up cake batter in a play kitchen.   While at play, an intensive language therapy program is also underway for preschool-aged children with developmental language disorder (DLD) at the University of Delaware’s Treatment Efficacy and Learning Language (TELL) Lab. The child says, “Him cooking.” Owen Van Horne repeats back, “He is cooking.”  DLD is a problem with learning and using language not attributed to a hearing impairment or intellectual disability, according to DLD ...

LAST 30 PRESS RELEASES:

‘Molecular shield’ placed in the nose may soon treat common hay fever trigger

Beetles under climate stress lay larger male eggs: Wolbachia infection drives adaptive reproduction strategy in response to rising temperature and CO₂

Groundbreaking quantum study puts wave-particle duality to work

Weekly injection could be life changing for Parkinson’s patients

Toxic metals linked to impaired growth in infants in Guatemala

Being consistently physically active in adulthood linked to 30–40% lower risk of death

Nerve pain drug gabapentin linked to increased dementia, cognitive impairment risks

Children’s social care involvement common to nearly third of UK mums who died during perinatal period

‘Support, not judgement’: Study explores links between children’s social care involvement and maternal deaths

Ethnic minority and poorer children more likely to die in intensive care

Major progress in fertility preservation after treatment for cancer of the lymphatic system

Fewer complications after additional ultrasound in pregnant women who feel less fetal movement

Environmental impact of common pesticides seriously underestimated

The Milky Way could be teeming with more satellite galaxies than previously thought

New study reveals surprising reproductive secrets of a cricket-hunting parasitoid fly

Media Tip Sheet: Symposia at ESA2025

NSF CAREER Award will power UVA engineer’s research to improve drug purification

Tiny parasitoid flies show how early-life competition shapes adult success

New coating for glass promises energy-saving windows

Green spaces boost children’s cognitive skills and strengthen family well-being

Ancient trees dying faster than expected in Eastern Oregon

Study findings help hone precision of proven CVD risk tool

Most patients with advanced melanoma who received pre-surgical immunotherapy remain alive and disease free four years later

Introducing BioEmu: A generative AI Model that enables high-speed and accurate prediction of protein structural ensembles

Replacing mutated microglia with healthy microglia halts progression of genetic neurological disease in mice and humans

New research shows how tropical plants manage rival insect tenants by giving them separate ‘flats’

Condo-style living helps keep the peace inside these ant plants

Climate change action could dramatically limit rising UK heatwave deaths

Annual heat-related deaths projected to increase significantly due to climate and population change

Researchers discover new way cells protect themselves from damage

[Press-News.org] To excel at engineering design, generative AI must learn to innovate, study finds
AI models that prioritize similarity falter when asked to design something completely new.