School of Humanities, Arts, and Social Sciences welcomes six new faculty for 2026

The MIT School of Humanities, Arts, and Social Sciences (SHASS) and Dean Agustín Rayo recently welcomed six new professors to the MIT community. They arrive with diverse backgrounds and vast knowledge in their areas of research.

Grisha Coleman is a full professor in the Music and Theater Arts Section. Her research explores tensions between our physiological, technological, and ecological systems; human movement, our machines, and the places we inhabit. Her practice engages an interdisciplinary approach to these explorations. Coleman received the Doris Duke Foundation’s Performing Arts Technologies Lab Award. Her work has been supported by Carnegie Mellon University’s STUDIO for Creative Inquiry, Creative Capital, the Jerome Foundation, MacDowell, the MAP Fund, the National Endowment for the Arts, the New York Foundation for the Arts, Pioneer Works, the Rockefeller Foundation Bellagio Center, Stanford University’s Mohr Visiting Artist program, and the Surdna Foundation. Coleman was previously a professor at Northeastern University and an associate professor at Arizona State University. She earned an MFA in music composition and integrated media from California Institute of the Arts.

Tung-Hui Hu is an associate professor with tenure in the Comparative Media Studies/Writing program. A poet and a scholar of digital media, he is the author of five books, most recently “Digital Lethargy: Dispatches from an Age of Disconnection” (MIT Press, 2022), “A Prehistory of the Cloud” (MIT Press, 2015), and “Greenhouses, Lighthouses” (Copper Canyon Press, 2013). Hu is interested in how concepts such as race and normal language became measurable, governable objects in the form of datasets. His research on data centers, artificial intelligence, burnout, and visual art has been featured in places such as CBS News, BBC Radio 4, WIRED, and MoMA R&D. He has been awarded fellowships from the American Academy in Rome, the National Endowment for the Arts, and the American Academy in Berlin. Prior to joining MIT, he was a faculty member at the University of Michigan.

Claire Luchette is an assistant professor in the Comparative Media Studies/Writing program. Luchette is the author of the novel “Agatha of Little Neon.” The winner of a Whiting Award and a National Book Foundation 5 Under 35 Honoree, Luchette has received fellowships from the Harvard Radcliffe Institute, the New York Public Library's Cullman Center for Scholars and Writers, MacDowell, Yaddo, and the National Endowment for the Arts. Their writing appears in Best American Short Stories, Ploughshares, and the Pushcart Prize anthology. Their second novel, “Swans,” and a story collection, “Big Whoop,” are forthcoming.

Shota Momma is an associate professor in the Department of Linguistics and Philosophy. Momma is a specialist in psycholinguistics and its interaction with linguistic theory — with a particular focus on the mechanisms of sentence production. Previously, Momma taught as an assistant professor at the University of Massachusetts Amherst. He earned a PhD in linguistics from the University of Maryland and completed a postdoctoral fellowship at the University of California San Diego. 

Lindsey Raymond PhD ’24 is an assistant professor in the Department of Economics, holding an MIT Schwarzman College of Computing shared position with the Department of Electrical Engineering and Computer Science. Her research examines how new technologies shape labor markets and market competition, and how insights from economics can inform algorithm design. She is a Schmidt Sciences AI2050 Early Career Fellow and served as a staff economist at the White House Council of Economic Advisers in 2021–22. Before joining MIT, Raymond was a postdoc at Microsoft Research. She earned her PhD from MIT and her BA from Yale University.

Makoto Harris Takao is the Class of 1957 Career Development Professor in the Music and Theater Arts Section. Working at the intersection of cultural history, religious studies, and musicology, Takao maps Japan’s entanglement with other world regions over the past 500 years. His current book project, “The Clef and the Cross: Music and Kirishitan Transculturation in Sixteenth-Century Japan,” asks what early modern Japanese Catholicism sounded like and how it was understood and expressed through Buddhist frameworks of sound, music, and movement. His work to date has appeared in such venues as Early Music, Journal of Music History Pedagogy, Journal of Religious History, Journal of Jesuit Studies, Zeithistorische Forschungen, and Oxford Bibliographies in Music. A player of the viola da gamba, Takao completed a joint PhD in history and musicology at the University of Western Australia. Before joining MIT, he was an assistant professor of musicology at the University of Illinois at Urbana-Champaign.

de MIT News https://ift.tt/nxgsz0L

Following the questions where they lead

Ever since she was a child playing on her family’s farmland in Wisconsin, Bailey Flanigan was guided by her own selective, yet wide-ranging, curiosity. Describing her young self as spirited and a bit unruly, she directed her energies to everything from building booby traps to doing experimental construction projects to exploring an intense interest in medicine to writing fiction and music to planning nonprofit organizations to help lessen social inequality.

By high school, Flanigan was intensely drawn to particular subjects.

“I found myself unmotivated to take all the AP [advanced placement] classes for the sake of it. My interest was captured by classes where I could be creative — where I could use math to solve real-world problems, creatively write, make music, connect distant ideas, or deeply explore the humanities — and I worked on such classes obsessively, as an opportunity to explore my intuitions and interests,” she says. “Instead of joining clubs, I ended up spending a lot of time thinking and creating on my own, and trying to understand what I enjoyed.”

Today Flanigan is a shared faculty member between the MIT Schwarzman College of Computing and the MIT departments of Political Science and Electrical Engineering and Computer Science (EECS), and a principal investigator in the MIT Laboratory for Information and Decision Systems. She has been involved in research at the University of Wisconsin, the National Institutes of Health, Google, and Carnegie Mellon, Drexel, Harvard, Princeton, and Stanford universities. Her current work focuses on using computational and mathematical tools to create new avenues for meaningful democratic participation.

Perhaps not surprisingly, her path has crossed huge expanses of subject matter and specialties — from medicine and bioengineering to public health, and from economics to her joint appointment at MIT in computer science and political science, which began in fall 2025.

“My trajectory across disciplines was just a result of me chasing down the problems I felt were most pressing or inspiring at the time. Along the way, I wound up in a lot of situations where I was less well-trained or qualified in the standard ways. While this was sometimes precarious, it was also incredibly fun, and it cultivated my ability to learn the languages of new disciplines more easily — a skill pretty much essential to my current research and job.”

In college at the University of Wisconsin at Madison, Flanigan worked in a wet lab on therapeutic targets in cancer and computationally on tumor genetics. She says she found the research intellectually interesting, but eventually began to wonder about whether it would have the kind of impact she wanted.

“At the time, I started to worry that the science I was developing might only, in the best case, be used by a small, relatively wealthy fraction of the world, when there were people suffering from much more-preventable diseases in much larger numbers,” she says.

So Flanigan moved toward public health, where she researched microfluidic devices for HIV detection that could be used in low-resource settings. Still bothered by the circumstances driving these settings’ limited resources to begin with, she then started to dabble in economics.

Around the same time, Flanigan’s academic advisors were chipping away at preconceptions she held about her own abilities.

Steven Wright, a professor of law and creative writing at UW-Madison, served as Flanigan’s informal mentor throughout college, and they worked together on a case at the Wisconsin Innocence Project.

“He guided me through my evolving interests in science, social inequality, and economics,” she says. “He was one of the people most responsible for convincing me that I could aim higher in my career, and that I could actually go to places like MIT or Harvard.”

Also while she was in college, the two heads of the UW-Madison scholarship office, Debbie Berger and Julie Stubbs, sent Flanigan repeated emails, encouraging her to apply for a Goldwater Scholarship.

“I kept deleting their emails, thinking they were spam — I didn’t think I was the kind of person that would apply for something like that. Their persistence convinced me to apply, and in the process, the horizons I perceived for myself started to change,” she says.

After graduating from UW-Madison, Flanigan worked as a predoctoral research assistant in economics at Princeton. There, Professor Evita Nestoridi, now an associate professor at Stony Brook University, also provided a pivotal moment of support, letting Flanigan audit her real analysis class.

“Evita’s class was my first real exposure to formal mathematics and proofs, and I loved it so much that it completely changed my career trajectory,” Flanigan says. “Despite my initial doubts, she convinced me that I could do math at the graduate level; because of her encouragement, I applied to computer science PhD programs the subsequent fall.”

Choosing Carnegie Mellon for her PhD, Flanigan began research on social choice and democratic decision-making, serving her dual passions for technical research and the issue of “who gets what and why,” she says, quoting Nobel Prize-winning economist Al Roth. 

Flanigan has developed algorithms that randomly choose participants of citizens’ assemblies, designed for the common case where willing participants self-select in ways that do not reflect the larger population. In a policy brief, Flanigan gave a hypothetical  example of an assembly on artificial intelligence, whose willing participants might skew toward younger, more educated citizens with an interest in technology, leaving other groups underrepresented despite their stake in the issue. The tools Flanigan has developed help balance representation with such features of the selection process as equality among individuals’ chances to participate, resistance to manipulation of the process, and transparency — all of which can affect the general perception of a decision-making group’s legitimacy.

Flanigan’s work is now deployed on panelot.org, a widely used open-access website hosting algorithms for randomly selecting citizen assembly participants.

“The site basically walks practitioners through a series of otherwise very technical trade-offs, making those trade-offs legible and then optimizing according to the priorities practitioners dictate,” she says.

Flanigan says she is motivated to improve how the public makes political decisions, “because if any political solution is going to be viable, the public needs to feel that it was arrived at via a legitimate political process — at least under the forms of government I find most appealing.”

Beyond her work on citizens’ assemblies, Flanigan’s research is exploring new avenues related to how to more systematically get public input on complex decisions, and how the format of questions we ask people in preference elicitation contexts can affect the substance of what we conclude.

“I feel so lucky to be studying these questions from within both political science and EECS, because I have the freedom to explore both the political and technical substance of tools for more direct governance as deeply as I want,” she says.

Flanigan’s curiosity-driven journey through widely varying terrain feels right in the MIT environment, she says.

“From the beginning, I got this sense of belonging at MIT — like my ways of thinking and problem-solving, which had seemed peculiar in many situations, actually made me belong more,” she says. “This was a super refreshing feeling, and it has been 100 percent borne out since I arrived.”

de MIT News https://ift.tt/5KrXUS4

How an influx of salt may affect microbial ecosystems

As sea levels rise due to climate change, encroaching sea water will likely make freshwater environments saltier. In a new study, MIT researchers have shown how that increase in salinity might affect microbial ecosystems found in environments such as rivers and estuaries.

These microbial communities play important roles in the carbon cycle, and they also help to decompose organic matter such as algae. The MIT team found that when salt levels rise, these populations lose diversity as faster-growing strains tend to take over the community, but they maintain their overall growth rate.

“At higher salinity, you lose diversity, which is ultimately not good for an ecosystem. But what we were surprised at is that in the meantime, even though diversity decreases, the growth of the community and the production of biomass is not impacted that much,” says Jana Huisman, an MIT postdoc and the lead author of the new study.

Jeff Gore, an MIT professor of physics, is the senior author of the paper, which appears today in Nature Microbiology. Martina Dal Bello, a former MIT postdoc who is now an assistant professor of ecology and evolutionary biology at Yale University, is also an author of the study.

Rising salt levels

Microbes that live in aquatic environments are typically adapted to thrive in fresh or salt water, or somewhere in between. Microbes that live in higher salt environments have cell walls that are optimized to resist osmotic pressure, and membrane transporters that can pump sodium ions out of the cell.

Freshwater lakes and rivers have salt concentrations around 1 gram of salt per liter of water (g/L), while oceans can reach 35 g/L. As the climate warms and sea levels rise, those oceanic waters may seep into estuaries and other inland bodies of water, increasing their salinity.

“When you think about climate change, you can think about rising temperatures, which is very common, but also a lot of other environmental stresses are going to increase,” Huisman says.

Huisman is from the Netherlands, a country with an extensive coastal delta, and she was interested in exploring how changes in salinity might affect microbial ecosystems in those aquatic habitats. The new study builds on previous work from Gore’s lab showing that higher seawater temperatures tend to favor slower-growing bacteria. 

For the new study, the researchers took samples from three aquatic environments with varying salinity: the Charles River near the MIT Sailing Pavilion (4 g/L), Boston Harbor (30 g/L), and a beach in Nahant, Massachusetts (35 g/L). Each community contained hundreds of species of microbes. The researchers then grew each population in three environments of varying salinity — 16, 31, or 46 g/L.

Over two weeks, the researchers measured the communities’ growth rates and found that overall, each community maintained the same growth rate at each of the three concentrations. However, in the communities exposed to higher salt environments, the overall composition became less diverse. Further studies showed that these communities tended to be dominated by faster-growing species. 

“We saw that those communities that had been propagated at higher salinity had reached a markedly different composition than the ones that lower salinity,” Huisman says.

Natural ecosystems

To explore whether their lab results might correspond to what happens in natural ecosystems, the researchers analyzed publicly available genomic data from microbes found in different aquatic ecosystems, including the Chesapeake Bay, the Gulf of Mexico, and the Baltic Sea.

For this portion of the study, the researchers focused on a genetic marker called the 16S rRNA gene copy number, which can be used as a proxy for the maximum growth rate that a species can attain. The more copies of this gene that a species has, the faster its intrinsic growth rate.

The researchers found that in these natural communities, environments with higher salinity also tended to be dominated by faster-growing species.

“When we first saw that, it was very exciting — that, indeed, what we found in the lab seems to also be represented in data from natural communities, sampled across a range of different environments,” Huisman says. “You see the same signatures in such data, and that’s highly suggestive that what we found in the lab might also be true in natural environments.”

One potential drawback to this loss of diversity is a reduction in microbial populations’ ability to withstand other types of environmental stress, the researchers say.

In this study, the researchers did not investigate the functions of the individual bacterial strains that ended up becoming more prevalent. Some of them may play beneficial roles, but it’s also possible that some of them might be pathogenic strains.

“Whether you want faster-growing species to take over or not might also be related to what the identity of those species is. That is something that I’m interested in looking at in the future,” Huisman says. 

The research was funded by a Human Frontier Science Program Fellowship and a Schmidt Science Polymath Award.

de MIT News https://ift.tt/vu1bw45

Diana Grass: Listening to the body’s language

Growing up in Colombia, Diana Grass had a simple response whenever someone told her something was impossible.

“I’ll figure it out.”

It’s a phrase that she still lives by today as PhD candidate in the Harvard-MIT Program in Health Sciences and Technology (HST), as she develops soft bioelectronic devices to study the physiological signals through which the brain and body communicate. 

“I’ve always been fascinated by one question: How do complex systems work?” Grass says.

An instinct to get to the bottom of things has guided Grass’ unconventional academic journey across continents and disciplines. Before becoming a neuroscientist and engineer, Grass studied philology and education to understand how language evolves, preserves knowledge, and shapes human communication. Looking back, she sees a common thread. “I wasn’t just studying language,” she says. “I was learning how complex systems communicate.”

But it wasn’t until she moved to the United States and began working as a medical interpreter that her scientific interests took a new direction.

“Every day, I translated conversations between physicians and patients with neurological disorders,” she says. “Watching those interactions sparked a fascination with the brain. I was intrigued by how a single organ could shape how we communicate, and ultimately who we are.”

Working alongside clinicians, Grass watched them rely on laboratory tests, medical imaging, and vital signs to understand what was happening inside the body. Despite remarkable advances in medical imaging and diagnostics, clinicians still rely largely on isolated snapshots of biological processes that are continuously changing inside the body.

“The body is communicating all the time,” she says. “We still lack the tools to understand its language.”

Determined to better understand the brain, Grass returned to school to study neuroscience with a minor in pre-medicine. She joined an immunology laboratory at Rutgers New Jersey Medical School, where she investigated neuroimmune communication and gained a new appreciation for the body’s interconnected physiology.

“Until then, I had been fascinated by the brain,” she says. “My work in immunology made me realize that the nervous system doesn’t function in isolation,” she says. “It continuously communicates with the immune system and peripheral organs to coordinate physiology and maintain homeostasis. To understand health and disease, we have to understand how those interactions preserve or disrupt that balance.” 

That realization transformed her scientific focus from understanding the brain to understanding how the nervous system coordinates physiology through continuous communication with the rest of the body, beginning with the immune system.

The complexity of that question ultimately brought Grass to pursue a PhD in medical engineering and medical physics with the HST program. She works in the Bioelectronics Group, led by Polina Anikeeva, the Matoula S. Salapatas Professor and head of MIT’s Department of Materials Science and Engineering, and also uses facilities in the T.J. Rodgers Laboratory and MIT.nano.

Today, Grass develops soft bioelectronic devices that integrate seamlessly with soft peripheral tissues without damaging them, to continuously monitor multiple physiological signals while enabling electrical recording and stimulation of neural circuits. These technologies provide a new way to investigate how neural communication coordinates physiology across the entire body. This knowledge could enable earlier diagnosis, more precise therapies, and a new generation of bioelectronic medicine.

For Grass, the work has taken on an even deeper significance since becoming a mother. Grass has two school-age children and for her, the possibility of developing technologies that help detect disease earlier and personalize treatments isn’t just a scientific goal; it’s one she hopes will shape the future of medicine for the next generation. 

“I want to contribute to a future where medicine understands the body physiology well enough to predict disease instead of simply reacting to it, personalize therapies with greater precision, and ultimately give families more healthy years together,” she says. “Because once you become a parent, every scientific question becomes deeply human.”

The complexity of Grass’ research has required her to step well beyond her original training. After studying neuroscience and immunology, she immersed herself in materials science, systems physiology, device fabrication, bioelectronics, and surgery to develop the tools needed to answer fundamental biological questions.

“The scientific question was bigger than any one discipline,” she says. “HST taught me to begin with biology, not disciplines. Once you understand the biological principles, medicine, engineering, and science stop being separate fields. They become complementary ways of answering the same question.”

The constant need to learn a new discipline has been both the most rewarding and challenging part of Grass’ research so far. 

“Every time I crossed into a new discipline, I felt like an immigrant again,” she says. “I had to learn a new language, understand a new culture, and earn the trust of people who had spent their careers there.”

Grass’ passion for understanding cultures extends well beyond the lab. Soon after arriving at MIT, she co-founded the Graduate First-Generation Low-Income Student Group to create a supportive space for students and connect them with the resources they need to thrive. What began as a small initiative has grown into a community of more than 300 graduate students representing over 60 countries, connecting students with faculty, alumni, entrepreneurs, and industry leaders.

“It has been really rewarding to see new GFLI leaders emerge and continue this legacy,” Grass says.

As an avid traveler, Grass’ favorite pastime is exploring new cultures, whether that be through learning a new traditional recipe or a new language. She speaks four languages fluently and can say “thank you” in roughly 50 more.

Whether she’s cooking Thai food with her children or introducing friends to recipes from around the world, she sees food as another language capable of connecting people across cultures. That same philosophy shapes how she thinks about science.

“I’ve realized that every culture has its own language and every scientific discipline its own way of understanding the world,” she says. “Looking back, every stage of my life has been about understanding how complex systems communicate. Today, my goal is to help medicine understand the principles that govern communication across the human body in health and disease.”

de MIT News https://ift.tt/4Ux1CIL

For energy systems that power a reliable grid, the future is all about location

Will a warming climate and changing weather patterns lead to more grid blackouts and other energy disruptions? Answering that question requires studying both regional climate forecasts and local energy systems, including emerging renewable generation, storage, transmission lines, and demand forecasts. The lack of such studies is one reason why energy developers and grid operators rarely consider climate change when deciding where to build their next project.

Now MIT researchers have created a way to make more climate-informed energy siting choices, and shown how it can be used to make energy systems more resilient and reduce blackouts. The researchers’ framework, described today in Nature Energy, combines fine-scale meteorology with detailed simulations of energy infrastructure. It shows how the location of new energy projects will play a significant role in meeting future demand in a changing climate.

The researchers applied their framework to decarbonized energy systems in New England and Texas, finding that energy systems designed for historic climate conditions could face up to a fivefold increase in energy shortfalls, potentially leading to blackouts, by 2050. Taking climate change into account when designing the system, conversely, improved the resilience of both regions’ energy systems at no or very little additional costs.

“As we mitigate climate change with renewables, we can also adapt to climate change by using future weather projections in our power system planning, and the extra costs of that adaptation are, at least in this study, not much,” says senior author Michael Howland, MIT’s Jeffrey Cheah Career Development Professor. “It’s different from other climate adaptation studies, where building a big seawall or other mitigation efforts are really expensive. In this case, if we’re smart when we design our power system decarbonization plans, it could cost almost nothing extra to simultaneously adapt to climate change.”

Joining Howland on the paper are first author Liying Qiu, a former MIT postdoc; Rahman Khorramfar and Shen Wang, current postdocs at MIT; and Saurabh Amin, MIT’s Edmund K. Turner Professor in Civil Engineering.

A better way to think about energy projects

The world’s energy systems are in a period of change. On the demand side, that change is driven by trends like the rising demand for artificial intelligence and the electrification of industries including transportation. On the supply side, that change is driven by the plummeting costs of renewable systems like solar and wind energy.

“That drop in costs has enabled the widespread deployment of renewables, because they’re the cheapest electricity-generation solution in many locations,” Howland explains. “At the same time, for the first time in more than a decade, electricity demand is starting to increase in the U.S.”

As low-cost variable renewable energy supplies increase, matching supply and demand throughout the day can become a harder problem for energy system operators. Adding to that complexity is the fact that renewables and energy demand are both influenced by weather and climate in different ways in different regions.

In the past, researchers have generally studied the impacts of climate change on individual technologies, for instance studying how it might change global wind and solar patterns. Other studies have considered the impact of climate change on states or other large areas, overlooking the specifics of regional energy systems. More recently, region-specific studies have been done but typically relied on low-resolution, global climate models.

“That’s what climate models are good at: giving you the global picture at coarse resolution,” Howland explains. “That limits insights for regional system planning and risk assessments.”

For their paper, the MIT researchers chose to study Texas and New England because they provided two different climate types and energy systems. The team used fine-scale meteorology models and considered the influence of climate change on weather-related energy failures.

“This study explores the joint, simultaneous impacts on multiple components of the energy system, similar to compound events studied in climate science,” Howland explains. “An extreme weather event can impact wind and solar generation and electricity demand all at the same time. Our hypothesis is that’s likely to be the biggest impact we’ll see from climate change on energy systems.”

The researchers also considered the impact of using climate change models to help site energy projects, looking out to 2050 because that’s the typical lifetime of wind and solar plants being built today. They found that locations that are best suited to provide the renewable wind and solar energy that the grid needs were meaningfully different in future climate conditions than in the historic climate.

The researchers found that climate change could increase energy failures by as much as 500 percent by 2050 if the siting did not consider future climate conditions. Such failures were driven primarily by the interaction between multiday renewable shortfalls and energy system design decisions like where to build solar farms and transmission lines.

“We are telling people where you put your wind and solar matters a lot for your ability to deliver energy when you need it,” Qiu explains. “We need to think more about the when and where of adding renewables rather than only focusing on adding overall capacity.”

In New England’s power system, the researchers found that energy supply disruptions caused by climate-related weather changes necessitate investment in solar capacity and transmission lines close to energy demand centers like cities. In Texas, energy disruption risks were primarily driven by transmission constraints.

The researchers found that climate-informed designs would prioritize adding wind farms in West Texas to better align with future demand patterns. The study assumes both regions will continue adding renewable capacity, thus the researchers concluded that Texas could improve the resilience of its grid at near-zero additional cost.

“We are showing that increasing energy resilience requires more than just spending more money,” Qiu says. “It primarily requires better and smarter planning.”

A new approach to adaptation

Howland says taking a broader view of climate change’s impact on energy systems helped his team get a clearer picture of blackout risks and other potential supply problems.

“On the individual power plant level, it’s not necessarily that climate change is a dominant uncertainty, so it really comes down to how all these energy system components and energy demand relate to each other,” Howland says. “That’s where we see the biggest impact of climate change, rather than on the level of individual wind or solar plants.”

Because the researchers used expensive, high-resolution models, Howland says their new model wouldn’t be practical for grid operators to use in their daily work today, but they hope to soon develop faster models that grid operators could use more easily.

“This study shows the opportunity and the need,” Howland says. “There are risks to not adapting our system, but if we do adapt our system, there could be big opportunities that are not costly. Now the key challenge is that we have to address the massive data and translation gap we have between meteorology and energy system planning and management. Right now, there’s too big of a divide between climate and weather modelers and power system practitioners. We want to continue to break that barrier down through interdisciplinary research.”

This work was supported by the MIT Climate Grand Challenges, the MIT Climate and Sustainability Consortium, and the MIT Energy Initiative Future Energy Systems Center.

de MIT News https://ift.tt/n1F3u9c

A better way to turn 2D designs into 3D models for rapid prototyping

Engineers often use vision-language models to produce new designs, such as for airplane or automobile components. To simulate how those components will perform in realistic situations, they’ll use tried-and-true computer-aided design (CAD) software to generate 3D models of those designs, which they can put through virtual crash or durability tests. 

Researchers from MIT and elsewhere have now developed a system that can teach a vision-language model to automatically convert 2D designs into CAD programs that are much more accurate and functional compared to other approaches, while using only a fraction of the computation.

By improving the performance and efficiency of AI-driven CAD generation, this technique could streamline the rapid prototyping process and reduce costs. It could also help engineers identify beneficial design choices they might otherwise overlook. 

The system generates new data based on the model’s abilities as it attempts to convert a 2D image into a CAD program. The framework corrects the model’s failures and incorporates them into a dataset with its successful solutions. 

It uses these data to teach the model how to fix specific mistakes and tackle tricky problems it would struggle with on its own.

“We want engineers to be able to point our framework at an underperforming CAD model, set a compute budget, and let the system take over — turning the model’s own mistakes into better training data,” says lead author Giorgio Giannone, a research affiliate in the Design Computation and Digital Engineering (DeCoDE) Lab at MIT and a principal research scientist on the AI Innovation Team at Red Hat.

He is joined on the paper by Anna Claire Doris, a mechanical engineering graduate student at MIT; Amin Heyrani Nobari, an MIT postdoc; Kai Xu of RedHat; and co-senior authors Akash Srivastava, director of Core AI at IBM and a principal investigator at the MIT-IBM Computing Research Lab; and Faez Ahmed, associate professor of mechanical engineering at MIT, leader of the DeCoDE Lab, and a principal investigator at the MIT-IBM Computing Research Lab. The research was recently presented at the International Conference on Machine Learning.

“Nearly every physical product around us, from airplanes to appliances, begins its life as a CAD model. Industry teams are eager for AI that can help speed-up the creation of these designs, but today's models often produce simple shapes inadequate for practice. What excites me about this work is that it gives many image-to-CAD-code models a way to improve themselves, learning from their own errors rather than waiting for more human-made data — and that brings trustworthy AI design tools much closer to everyday engineering,” says Ahmed.

Model-aware data

The researchers are working toward building vision-language models (VLMs) for CAD generation. These VLMs take a 2D image and some descriptive text, and output Python code that can be executed in a CAD software program to generate a 3D model of a physical object.

They studied the challenges of deploying existing VLMs for this task and determined the main bottleneck that limits their capabilities is the lack of diverse, high-quality CAD datasets to train them. 

To remedy this, they sought to create new data to teach a model how to perform CAD generation, using a process known as data augmentation.

In data augmentation, scientists typically create new data by randomly tweaking existing data to generate more samples, often by adjusting the color, size, and shape of objects in images. 

Instead, the MIT researchers built a data augmentation system called GIFT (which stands for Geometric Inference Feedback Tuning) that generates data designed to improve the performance of one VLM for a specific task.

GIFT develops an understanding of the model’s strengths and weaknesses by testing it. Then it uses this knowledge to generate data that could improve the model’s performance on the CAD generation problems it struggles to solve.

“We want to obtain data augmentation that is informed by the model itself,” Giannone says. 

Learning from mistakes

To do this, GIFT asks the model to generate code that solves a CAD generation problem multiple times in parallel. It checks the correctness of these guesses to understand how well the model can solve this problem.

“For a model, generating CAD query code that is almost correct is not that hard, but generating code that is perfectly correct and can be executed is much more challenging for a standard VLM,” Giannone says.

For guesses that are nearly correct, GIFT adjusts them to become successful solutions. It saves these “near-misses” and successful solutions in a new dataset that can teach the model how to overcome problems that would usually trip it up.

“If we sample the model 10 times and it generates 10 correct answers to the same problem, then there is not much for it to learn. We care about the in-between cases, where the model might only solve the problem 50 percent of the time,” he says.

Using these in-between cases allows GIFT to generate data augmentations that are both model-aware and task-aware. In addition, by incorporating multiple correct solutions to the same problem, the new data expand the model’s general knowledge of CAD code generation.

This automatic system does not require human intervention to correct the model’s mistakes.

GIFT creates data augmentations from a pre-trained VLM using a process known as inference-time scaling. This process allows a static model, which has already been trained, to generate better outputs without the high computational costs of retraining the entire model. 

Using inference-time scaling, the user can determine how much computation they want to use for GIFT, tailoring it to their time and budget constraints. 

GIFT outperformed several competing techniques, generating CAD programs that were more accurate while using only about 20 percent as much computation. The CAD models generated by VLMs using GIFT were better aligned with the shapes of ground-truth models.

“With GIFT, we started with geometry because with engineering problems, if the geometry of a 3D shape is not correct, nothing else will be correct, but there are many other aspects to consider,” Giannone says.

In the future, the researchers want to expand GIFT so the framework can teach models to generate CAD programs that improve the performance and manufacturability of 3D models. They also want to apply the system to larger models and more diverse CAD generation tasks.

This research was funded, in part, by the MIT-IBM Computing Research Lab. 

de MIT News https://ift.tt/r1NXadk

MIT Professor Susumu Tonegawa, renowned molecular biologist and Nobel laureate, dies at 86

Susumu Tonegawa, the Picower Professor of Biology and Neuroscience at MIT and a Nobel laureate, died July 11 at the age of 86.

Tonegawa was a renowned molecular biologist who wielded his keen insight in a variety of fields, including immunology and neuroscience. In the early 1980s, Tonegawa discovered how the immune system generates its incredible diversity of antibodies — a breakthrough that earned him the Nobel Prize in Physiology or Medicine in 1987.

Following that landmark achievement, he turned his attention to neuroscience, where his work has helped to reveal how the brain stores memories as traces called “engrams.”

An MIT faculty member for more than 40 years, Tonegawa also served as the founding director of MIT’s Picower Institute for Learning and Memory and director of the RIKEN Brain Science Institute of Japan, and was a Howard Hughes Medical Institute Investigator.

“Few scientists have reshaped our understanding of biology as profoundly as Susumu Tonegawa,” says Myriam Heiman, director of the Picower Institute. “His intellectual fearlessness, extraordinary creativity, and relentless pursuit of fundamental questions opened entirely new frontiers in both immunology and neuroscience. His influence on science and on the people who had the privilege of working alongside him is immeasurable.” 

Drawn to molecular biology

Born in Nagoya, Japan, Tonegawa spent his early years moving between rural towns, due to his father’s job as an engineer for a textile company. When it was time for him to go to high school, his parents sent him to a school in Tokyo, where he became interested in chemistry.

He was admitted to the University of Kyoto to study chemistry, and while there, he was drawn to the nascent field of molecular biology. He began his graduate studies at the Institute for Virus Research at the University of Kyoto, but after only a couple of months, his advisor, Professor Itaru Watanabe, suggested that he apply to a school in the United States, which had more advanced molecular biology programs.

Tonegawa took that advice and was accepted at the University of California at San Diego, where he studied how a virus called phage lambda controls gene transcription. After earning his PhD in 1968, he went on to a postdoc in a lab at the Salk Institute.

In that lab, Tonegawa began studying gene expression of a virus known as SV40. However, his U.S. visa was set to expire at the end of 1970, so he soon headed for a position at the newly established Basel Institute for Immunology in Switzerland. 

At the time, Tonegawa had little background in immunology, but he soon became fascinated by the 100-year-old question of “antibody diversity” — how the body’s immune system is able to generate hundreds of millions of antibodies from a relatively small set of genes. (The entire human genome contains about 20,000 genes.) That antibody diversity is what allows the immune system to recognize so many pathogens, including those it has never seen before.

With colleagues in Basel, Tonegawa discovered that each antibody protein is not encoded by its own gene — instead, genes for different components of the antibody can be randomly recombined to generate limitless combinations.

In 1987, Tonegawa was a solo recipient of the Nobel Prize for discovering that process, known as V(D)J gene rearrangement. In announcing the award, the Nobel committee noted that Tonegawa’s discoveries “explain the genetic background allowing the enormous richness of variation amongst antibodies. Beyond deeper knowledge of the basic structure of the immune system these discoveries will have importance in improving immunological therapy of different kinds, such as for instance the enforcement of vaccinations and inhibition of reactions during transplantation.”

From antibodies to engrams

In the early 1980s, after his groundbreaking antibody discoveries, Tonegawa began to feel the urge to turn to new research directions. He also wanted to return to the United States, so in 1981, he accepted the offer of a professorship at MIT’s Center for Cancer Research (today known as the Koch Institute for Integrative Cancer Research). There, he began working on T cells and contributed to scientists’ understanding of how T cells are able to generate a large diversity of T-cell receptors.

While at the CCR, he also began to study questions in neuroscience. As he told an interviewer from the Picower Institute in 2022, he was always in search of new scientific endeavors to keep him interested in his work.

“When I decided to become a scientist, my criteria of what to do was whether the scientific problem I got to solve was interesting or not. Whether I’m curious our not. I didn’t think about other things like, Could it be too risky? Can I really develop my career by venturing into the field I am not familiar with? That never occurred to me. I just followed my curiosity and instinct,” Tonegawa said in an interview published in the summer 2022 Picower Institute newsletter.

In 1994, he was chosen as the founding director for MIT’s Center for Learning and Memory, which became the Picower Institute for Learning and Memory in 2002. Tonegawa continued to serve as the center’s director until the end of 2006. 

Professor Li-Huei Tsai, who succeeded Tonegawa as the Picower Institute’s director, calls working alongside Tonegawa “one of the greatest honors of my career.”

“His passion, boundless energy, and unwavering pursuit of the fundamental mechanisms underlying memory were contagious, inspiring generations of neuroscientists to join and advance the field. Today, we lost a giant. His scientific legacy will continue to shape neuroscience for years to come, and he will be deeply missed by all of us,” she says.

Over the past two decades, Tonegawa’s lab has made significant discoveries in the field of memory research. In 2013, he and his colleagues reported that they had identified “engrams” in the brain’s hippocampus. These engrams consist of episodic memories — memories of experiences — that are stored in specific groups of hippocampal cells. Engrams encode elements including objects, space, and time, linked to a specific experience.

At that time, the researchers also found that it was possible to implant “false memories” in mice by using optogenetics to reactivate an existing engram while the animals formed a new memory. This prompted the mice to associate a new location with the memory of an event that had actually happened in a different location.

Later work from Tonegawa’s lab showed that engrams extend beyond the hippocampus and are stored across a widely distributed complex that spans many brain circuits. More recently, he had been working on engrams of “knowledge memory” to decipher the fundamental mechanism of abstract memory. His recent work also delved into how the emotional associations of memories are encoded, and how the brain maintains a timeline of chronological events.

In addition to the Nobel Prize, Tonegawa received many other awards, including the Albert and Mary Lasker Award for Basic Research in 1987, the Bristol-Myers Award for Distinguished Achievement in Cancer Research in 1986, and the David M. Bonner Lifetime Achievement Award from the University of California at San Diego in 2010. He was also known for training many scientists who are now leaders in the field of neuroscience.

Tonegawa was a longtime fan of the Boston Red Sox, and in May 2004, he had the opportunity to throw out the ceremonial first pitch at Fenway Park, as part of the team’s tribute to the Boston area’s scientific and medical communities.

He is survived by his wife, Mayumi Tonegawa ’92, two children, Hidde Tonegawa ’09 and Hanna Tonegawa, and two grandchildren. He was predeceased by a son, Satto Tonegawa. 

Following a private funeral, his ashes will be buried in Kyoto, Japan.

de MIT News https://ift.tt/T4la9VY