Posts in category: MIT

Human languages are notoriously complex, and linguists have long thought it would be impossible to teach a machine how to analyze speech sounds and word structures in the way human investigators do.

But researchers at MIT, Cornell University, and McGill University have taken a step in this direction. They have demonstrated an artificial intelligence system that can learn the rules and patterns of human languages on its own.

When given words and examples of how those words change to express different grammatical functions (like tense, case, or gender) in one language, this machine-learning model comes up with rules that explain why the forms of those words change. For instance, it might learn that the letter “a” must be added to end of a word to make the masculine form feminine in Serbo-Croatian.

This model can also automatically learn higher-level language patterns that can apply to many languages, enabling it to achieve better results.

The researchers trained and tested the model using problems from linguistics textbooks that featured 58 different languages. Each problem had a set of words and corresponding word-form changes. The model was able to come up with a correct set of rules to describe those word-form changes for 60 percent of the problems.

This system could be used to study language hypotheses and investigate subtle similarities in the way diverse languages transform words. It is especially unique because the system discovers models that can be readily understood by humans, and it acquires these models from small amounts of data, such as a few dozen words. And instead of using one massive dataset for a single task, the system utilizes many small datasets, which is closer to how scientists propose hypotheses — they look at multiple related datasets and come up with models to explain phenomena across those datasets.

“One of the motivations of this work was our desire to study systems that learn models of datasets that is represented in a way that humans can understand. Instead of learning weights, can the model learn expressions or rules? And we wanted to see if we could build this system so it would learn on a whole battery of interrelated datasets, to make the system learn a little bit about how to better model each one,” says Kevin Ellis ’14, PhD ’20, an assistant professor of computer science at Cornell University and lead author of the paper.

Joining Ellis on the paper are MIT faculty members Adam Albright, a professor of linguistics; Armando Solar-Lezama, a professor and associate director of the Computer Science and Artificial Intelligence Laboratory (CSAIL); and Joshua B. Tenenbaum, the Paul E. Newton Career Development Professor of Cognitive Science and Computation in the Department of Brain and Cognitive Sciences and a member of CSAIL; as well as senior author

Timothy J. O’Donnell, assistant professor in the Department of Linguistics at McGill University, and Canada CIFAR AI Chair at the Mila – Quebec Artificial Intelligence Institute.

The research is published today in Nature Communications.

Looking at language 

In their quest to develop an AI system that could automatically learn a model from multiple related datasets, the researchers chose to explore the interaction of phonology (the study of sound patterns) and morphology (the study of word structure).

Data from linguistics textbooks offered an ideal testbed because many languages share core features, and textbook problems showcase specific linguistic phenomena. Textbook problems can also be solved by college students in a fairly straightforward way, but those students typically have prior knowledge about phonology from past lessons they use to reason about new problems.

Ellis, who earned his PhD at MIT and was jointly advised by Tenenbaum and Solar-Lezama, first learned about morphology and phonology in an MIT class co-taught by O’Donnell, who was a postdoc at the time, and Albright.

“Linguists have thought that in order to really understand the rules of a human language, to empathize with what it is that makes the system tick, you have to be human. We wanted to see if we can emulate the kinds of knowledge and reasoning that humans (linguists) bring to the task,” says Albright.

To build a model that could learn a set of rules for assembling words, which is called a grammar, the researchers used a machine-learning technique known as Bayesian Program Learning. With this technique, the model solves a problem by writing a computer program.

In this case, the program is the grammar the model thinks is the most likely explanation of the words and meanings in a linguistics problem. They built the model using Sketch, a popular program synthesizer which was developed at MIT by Solar-Lezama.

But Sketch can take a lot of time to reason about the most likely program. To get around this, the researchers had the model work one piece at a time, writing a small program to explain some data, then writing a larger program that modifies that small program to cover more data, and so on.

They also designed the model so it learns what “good” programs tend to look like. For instance, it might learn some general rules on simple Russian problems that it would apply to a more complex problem in Polish because the languages are similar. This makes it easier for the model to solve the Polish problem.

Tackling textbook problems

When they tested the model using 70 textbook problems, it was able to find a grammar that matched the entire set of words in the problem in 60 percent of cases, and correctly matched most of the word-form changes in 79 percent of problems.

The researchers also tried pre-programming the model with some knowledge it “should” have learned if it was taking a linguistics course, and showed that it could solve all problems better.

“One challenge of this work was figuring out whether what the model was doing was reasonable. This isn’t a situation where there is one number that is the single right answer. There is a range of possible solutions which you might accept as right, close to right, etc.,” Albright says.

The model often came up with unexpected solutions. In one instance, it discovered the expected answer to a Polish language problem, but also another correct answer that exploited a mistake in the textbook. This shows that the model could “debug” linguistics analyses, Ellis says.

The researchers also conducted tests that showed the model was able to learn some general templates of phonological rules that could be applied across all problems.

“One of the things that was most surprising is that we could learn across languages, but it didn’t seem to make a huge difference,” says Ellis. “That suggests two things. Maybe we need better methods for learning across problems. And maybe, if we can’t come up with those methods, this work can help us probe different ideas we have about what knowledge to share across problems.”

In the future, the researchers want to use their model to find unexpected solutions to problems in other domains. They could also apply the technique to more situations where higher-level knowledge can be applied across interrelated datasets. For instance, perhaps they could develop a system to infer differential equations from datasets on the motion of different objects, says Ellis.

“This work shows that we have some methods which can, to some extent, learn inductive biases. But I don’t think we’ve quite figured out, even for these textbook problems, the inductive bias that lets a linguist accept the plausible grammars and reject the ridiculous ones,” he adds.

“This work opens up many exciting venues for future research. I am particularly intrigued by the possibility that the approach explored by Ellis and colleagues (Bayesian Program Learning, BPL) might speak to how infants acquire language,” says T. Florian Jaeger, a professor of brain and cognitive sciences and computer science at the University of Rochester, who was not an author of this paper. “Future work might ask, for example, under what additional induction biases (assumptions about universal grammar) the BPL approach can successfully achieve human-like learning behavior on the type of data infants observe during language acquisition. I think it would be fascinating to see whether inductive biases that are even more abstract than those considered by Ellis and his team — such as biases originating in the limits of human information processing (e.g., memory constraints on dependency length or capacity limits in the amount of information that can be processed per time) — would be sufficient to induce some of the patterns observed in human languages.”

This work was funded, in part, by the Air Force Office of Scientific Research, the Center for Brains, Minds, and Machines, the MIT-IBM Watson AI Lab, the Natural Science and Engineering Research Council of Canada, the Fonds de Recherche du Québec – Société et Culture, the Canada CIFAR AI Chairs Program, the National Science Foundation (NSF), and an NSF graduate fellowship.

Read More

When the MIT Lincoln Laboratory Supercomputing Center (LLSC) unveiled its TX-GAIA supercomputer in 2019, it provided the MIT community a powerful new resource for applying artificial intelligence to their research. Anyone at MIT can submit a job to the system, which churns through trillions of operations per second to train models for diverse applications, such as spotting tumors in medical images, discovering new drugs, or modeling climate effects. But with this great power comes the great responsibility of managing and operating it in a sustainable manner — and the team is looking for ways to improve.

“We have these powerful computational tools that let researchers build intricate models to solve problems, but they can essentially be used as black boxes. What gets lost in there is whether we are actually using the hardware as effectively as we can,” says Siddharth Samsi, a research scientist in the LLSC. 

To gain insight into this challenge, the LLSC has been collecting detailed data on TX-GAIA usage over the past year. More than a million user jobs later, the team has released the dataset open source to the computing community.

Their goal is to empower computer scientists and data center operators to better understand avenues for data center optimization — an important task as processing needs continue to grow. They also see potential for leveraging AI in the data center itself, by using the data to develop models for predicting failure points, optimizing job scheduling, and improving energy efficiency. While cloud providers are actively working on optimizing their data centers, they do not often make their data or models available for the broader high-performance computing (HPC) community to leverage. The release of this dataset and associated code seeks to fill this space.

“Data centers are changing. We have an explosion of hardware platforms, the types of workloads are evolving, and the types of people who are using data centers is changing,” says Vijay Gadepally, a senior researcher at the LLSC. “Until now, there hasn’t been a great way to analyze the impact to data centers. We see this research and dataset as a big step toward coming up with a principled approach to understanding how these variables interact with each other and then applying AI for insights and improvements.”

Papers describing the dataset and potential applications have been accepted to a number of venues, including the IEEE International Symposium on High-Performance Computer Architecture, the IEEE International Parallel and Distributed Processing Symposium, the Annual Conference of the North American Chapter of the Association for Computational Linguistics, the IEEE High-Performance and Embedded Computing Conference, and International Conference for High Performance Computing, Networking, Storage and Analysis. 

Workload classification

Among the world’s TOP500 supercomputers, TX-GAIA combines traditional computing hardware (central processing units, or CPUs) with nearly 900 graphics processing unit (GPU) accelerators. These NVIDIA GPUs are specialized for deep learning, the class of AI that has given rise to speech recognition and computer vision.

The dataset covers CPU, GPU, and memory usage by job; scheduling logs; and physical monitoring data. Compared to similar datasets, such as those from Google and Microsoft, the LLSC dataset offers “labeled data, a variety of known AI workloads, and more detailed time series data compared with prior datasets. To our knowledge, it’s one of the most comprehensive and fine-grained datasets available,” Gadepally says. 

Notably, the team collected time-series data at an unprecedented level of detail: 100-millisecond intervals on every GPU and 10-second intervals on every CPU, as the machines processed more than 3,000 known deep-learning jobs. One of the first goals is to use this labeled dataset to characterize the workloads that different types of deep-learning jobs place on the system. This process would extract features that reveal differences in how the hardware processes natural language models versus image classification or materials design models, for example.   

The team has now launched the MIT Datacenter Challenge to mobilize this research. The challenge invites researchers to use AI techniques to identify with 95 percent accuracy the type of job that was run, using their labeled time-series data as ground truth.

Such insights could enable data centers to better match a user’s job request with the hardware best suited for it, potentially conserving energy and improving system performance. Classifying workloads could also allow operators to quickly notice discrepancies resulting from hardware failures, inefficient data access patterns, or unauthorized usage.

Too many choices

Today, the LLSC offers tools that let users submit their job and select the processors they want to use, “but it’s a lot of guesswork on the part of users,” Samsi says. “Somebody might want to use the latest GPU, but maybe their computation doesn’t actually need it and they could get just as impressive results on CPUs, or lower-powered machines.”

Professor Devesh Tiwari at Northeastern University is working with the LLSC team to develop techniques that can help users match their workloads to appropriate hardware. Tiwari explains that the emergence of different types of AI accelerators, GPUs, and CPUs has left users suffering from too many choices. Without the right tools to take advantage of this heterogeneity, they are missing out on the benefits: better performance, lower costs, and greater productivity.

“We are fixing this very capability gap — making users more productive and helping users do science better and faster without worrying about managing heterogeneous hardware,” says Tiwari. “My PhD student, Baolin Li, is building new capabilities and tools to help HPC users leverage heterogeneity near-optimally without user intervention, using techniques grounded in Bayesian optimization and other learning-based optimization methods. But, this is just the beginning. We are looking into ways to introduce heterogeneity in our data centers in a principled approach to help our users achieve the maximum advantage of heterogeneity autonomously and cost-effectively.”

Workload classification is the first of many problems to be posed through the Datacenter Challenge. Others include developing AI techniques to predict job failures, conserve energy, or create job scheduling approaches that improve data center cooling efficiencies.

Energy conservation 

To mobilize research into greener computing, the team is also planning to release an environmental dataset of TX-GAIA operations, containing rack temperature, power consumption, and other relevant data.

According to the researchers, huge opportunities exist to improve the power efficiency of HPC systems being used for AI processing. As one example, recent work in the LLSC determined that simple hardware tuning, such as limiting the amount of power an individual GPU can draw, could reduce the energy cost of training an AI model by 20 percent, with only modest increases in computing time. “This reduction translates to approximately an entire week’s worth of household energy for a mere three-hour time increase,” Gadepally says.

They have also been developing techniques to predict model accuracy, so that users can quickly terminate experiments that are unlikely to yield meaningful results, saving energy. The Datacenter Challenge will share relevant data to enable researchers to explore other opportunities to conserve energy.

The team expects that lessons learned from this research can be applied to the thousands of data centers operated by the U.S. Department of Defense. The U.S. Air Force is a sponsor of this work, which is being conducted under the USAF-MIT AI Accelerator.

Other collaborators include researchers at MIT Computer Science and Artificial Intelligence Laboratory (CSAIL). Professor Charles Leiserson’s Supertech Research Group is investigating performance-enhancing techniques for parallel computing, and research scientist Neil Thompson is designing studies on ways to nudge data center users toward climate-friendly behavior.

Samsi presented this work at the inaugural AI for Datacenter Optimization (ADOPT’22) workshop last spring as part of the IEEE International Parallel and Distributed Processing Symposium. The workshop officially introduced their Datacenter Challenge to the HPC community.

“We hope this research will allow us and others who run supercomputing centers to be more responsive to user needs while also reducing the energy consumption at the center level,” Samsi says.

Read More

Read More

Parkinson’s disease is notoriously difficult to diagnose as it relies primarily on the appearance of motor symptoms such as tremors, stiffness, and slowness, but these symptoms often appear several years after the disease onset. Now, Dina Katabi, the Thuan (1990) and Nicole Pham Professor in the Department of Electrical Engineering and Computer Science (EECS) at MIT and principal investigator at MIT Jameel Clinic, and her team have developed an artificial intelligence model that can detect Parkinson’s just from reading a person’s breathing patterns.

The tool in question is a neural network, a series of connected algorithms that mimic the way a human brain works, capable of assessing whether someone has Parkinson’s from their nocturnal breathing — i.e., breathing patterns that occur while sleeping. The neural network, which was trained by MIT PhD student Yuzhe Yang and postdoc Yuan Yuan, is also able to discern the severity of someone’s Parkinson’s disease and track the progression of their disease over time. 

Yang and Yuan are co-first authors on a new paper describing the work, published today in Nature Medicine. Katabi, who is also an affiliate of the MIT Computer Science and Artificial Intelligence Laboratory and director of the Center for Wireless Networks and Mobile Computing, is the senior author. They are joined by 12 colleagues from Rutgers University, the University of Rochester Medical Center, the Mayo Clinic, Massachusetts General Hospital, and the Boston University College of Health and Rehabilition.

Over the years, researchers have investigated the potential of detecting Parkinson’s using cerebrospinal fluid and neuroimaging, but such methods are invasive, costly, and require access to specialized medical centers, making them unsuitable for frequent testing that could otherwise provide early diagnosis or continuous tracking of disease progression.

The MIT researchers demonstrated that the artificial intelligence assessment of Parkinson’s can be done every night at home while the person is asleep and without touching their body. To do so, the team developed a device with the appearance of a home Wi-Fi router, but instead of providing internet access, the device emits radio signals, analyzes their reflections off the surrounding environment, and extracts the subject’s breathing patterns without any bodily contact. The breathing signal is then fed to the neural network to assess Parkinson’s in a passive manner, and there is zero effort needed from the patient and caregiver.

“A relationship between Parkinson’s and breathing was noted as early as 1817, in the work of Dr. James Parkinson. This motivated us to consider the potential of detecting the disease from one’s breathing without looking at movements,” Katabi says. “Some medical studies have shown that respiratory symptoms manifest years before motor symptoms, meaning that breathing attributes could be promising for risk assessment prior to Parkinson’s diagnosis.”

The fastest-growing neurological disease in the world, Parkinson’s is the second-most common neurological disorder, after Alzheimer’s disease. In the United States alone, it afflicts over 1 million people and has an annual economic burden of $51.9 billion. The research team’s device was tested on 7,687 individuals, including 757 Parkinson’s patients.

Katabi notes that the study has important implications for Parkinson’s drug development and clinical care. “In terms of drug development, the results can enable clinical trials with a significantly shorter duration and fewer participants, ultimately accelerating the development of new therapies. In terms of clinical care, the approach can help in the assessment of Parkinson’s patients in traditionally underserved communities, including those who live in rural areas and those with difficulty leaving home due to limited mobility or cognitive impairment,” she says.

“We’ve had no therapeutic breakthroughs this century, suggesting that our current approaches to evaluating new treatments is suboptimal,” says Ray Dorsey, a professor of neurology at the University of Rochester and Parkinson’s specialist who co-authored the paper. Dorsey adds that the study is likely one of the largest sleep studies ever conducted on Parkinson’s. “We have very limited information about manifestations of the disease in their natural environment and [Katabi’s] device allows you to get objective, real-world assessments of how people are doing at home. The analogy I like to draw [of current Parkinson’s assessments] is a street lamp at night, and what we see from the street lamp is a very small segment … [Katabi’s] entirely contactless sensor helps us illuminate the darkness.”

This research was performed in collaboration with the University of Rochester, Mayo Clinic, and Massachusetts General Hospital, and is sponsored by the National Institutes of Health, with partial support by the National Science Foundation and the Michael J. Fox Foundation.

Read More

MIT researchers have developed a method for 3D printing materials with tunable mechanical properties, that sense how they are moving and interacting with the environment. The researchers create these sensing structures using just one material and a single run on a 3D printer.

To accomplish this, the researchers began with 3D-printed lattice materials and incorporated networks of air-filled channels into the structure during the printing process. By measuring how the pressure changes within these channels when the structure is squeezed, bent, or stretched, engineers can receive feedback on how the material is moving.

The method opens opportunities for embedding sensors within architected materials, a class of materials whose mechanical properties are programmed through form and composition. Controlling the geometry of features in architected materials alters their mechanical properties, such as stiffness or toughness. For instance, in cellular structures like the lattices the researchers print, a denser network of cells makes a stiffer structure.

This technique could someday be used to create flexible soft robots with embedded sensors that enable the robots to understand their posture and movements. It might also be used to produce wearable smart devices that provide feedback on how a person is moving or interacting with their environment.

“The idea with this work is that we can take any material that can be 3D-printed and have a simple way to route channels throughout it so we can get sensorization with structure. And if you use really complex materials, then you can have motion, perception, and structure all in one,” says co-lead author Lillian Chin, a graduate student in the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL).

Joining Chin on the paper are co-lead author Ryan Truby, a former CSAIL postdoc who is now as assistant professor at Northwestern University; Annan Zhang, a CSAIL graduate student; and senior author Daniela Rus, the Andrew and Erna Viterbi Professor of Electrical Engineering and Computer Science and director of CSAIL. The paper is published today in Science Advances.

Architected materials

The researchers focused their efforts on lattices, a type of “architected material,” which exhibits customizable mechanical properties based solely on its geometry. For instance, changing the size or shape of cells in the lattice makes the material more or less flexible.

While architected materials can exhibit unique properties, integrating sensors within them is challenging given the materials’ often sparse, complex shapes. Placing sensors on the outside of the material is typically a simpler strategy than embedding sensors within the material. However, when sensors are placed on the outside, the feedback they provide may not provide a complete description of how the material is deforming or moving.

Instead, the researchers used 3D printing to incorporate air-filled channels directly into the struts that form the lattice. When the structure is moved or squeezed, those channels deform and the volume of air inside changes. The researchers can measure the corresponding change in pressure with an off-the-shelf pressure sensor, which gives feedback on how the material is deforming.

Because they are incorporated into the material, these “fluidic sensors” offer advantages over conventional sensor materials.

“Sensorizing” structures

The researchers incorporate channels into the structure using digital light processing 3D printing. In this method, the structure is drawn out of a pool of resin and hardened into a precise shape using projected light. An image is projected onto the wet resin and areas struck by the light are cured.

But as the process continues, the resin remains stuck inside the sensor channels. The researchers had to remove excess resin before it was cured, using a mix of pressurized air, vacuum, and intricate cleaning.

They used this process to create several lattice structures and demonstrated how the air-filled channels generated clear feedback when the structures were squeezed and bent.

“Importantly, we only use one material to 3D print our sensorized structures. We bypass the limitations of other multimaterial 3D printing and fabrication methods that are typically considered for patterning similar materials,” says Truby.

Building off these results, they also incorporated sensors into a new class of materials developed for motorized soft robots known as handed shearing auxetics, or HSAs. HSAs can be twisted and stretched simultaneously, which enables them to be used as effective soft robotic actuators. But they are difficult to “sensorize” because of their complex forms.

They 3D printed an HSA soft robot capable of several movements, including bending, twisting, and elongating. They ran the robot through a series of movements for more than 18 hours and used the sensor data to train a neural network that could accurately predict the robot’s motion. 

Chin was impressed by the results — the fluidic sensors were so accurate she had difficulty distinguishing between the signals the researchers sent to the motors and the data that came back from the sensors.

“Materials scientists have been working hard to optimize architected materials for functionality. This seems like a simple, yet really powerful idea to connect what those researchers have been doing with this realm of perception. As soon as we add sensing, then roboticists like me can come in and use this as an active material, not just a passive one,” she says.

“Sensorizing soft robots with continuous skin-like sensors has been an open challenge in the field. This new method provides accurate proprioceptive capabilities for soft robots and opens the door for exploring the world through touch,” says Rus.

In the future, the researchers look forward to finding new applications for this technique, such as creating novel human-machine interfaces or soft devices that have sensing capabilities within the internal structure. Chin is also interested in utilizing machine learning to push the boundaries of tactile sensing for robotics.

“The use of additive manufacturing for directly building robots is attractive. It allows for the complexity I believe is required for generally adaptive systems,” says Robert Shepherd, associate professor at the Sibley School of Mechanical and Aerospace Engineering at Cornell University, who was not involved with this work. “By using the same 3D printing process to build the form, mechanism, and sensing arrays, their process will significantly contribute to researcher’s aiming to build complex robots simply.”

This research was supported, in part, by the National Science Foundation, the Schmidt Science Fellows Program in partnership with the Rhodes Trust, an NSF Graduate Fellowship, and the Fannie and John Hertz Foundation.

Read More

Caspar Hare and Georgia Perakis have been appointed the new associate deans of the Social and Ethical Responsibilities of Computing (SERC), a cross-cutting initiative in the MIT Stephen A. Schwarzman College of Computing. Their new roles will take effect on Sept. 1.

“Infusing social and ethical aspects of computing in academic research and education is a critical component of the college mission,” says Daniel Huttenlocher, dean of the MIT Schwarzman College of Computing and the Henry Ellis Warren Professor of Electrical Engineering and Computer Science. “I look forward to working with Caspar and Georgia on continuing to develop and advance SERC and its reach across MIT. Their complementary backgrounds and their broad connections across MIT will be invaluable to this next chapter of SERC.”

Caspar Hare

Hare is a professor of philosophy in the Department of Linguistics and Philosophy. A member of the MIT faculty since 2003, his main interests are in ethics, metaphysics, and epistemology. The general theme of his recent work has been to bring ideas about practical rationality and metaphysics to bear on issues in normative ethics and epistemology. He is the author of two books: “On Myself, and Other, Less Important Subjects” (Princeton University Press 2009), about the metaphysics of perspective, and “The Limits of Kindness” (Oxford University Press 2013), about normative ethics.

Georgia Perakis

Perakis is the William F. Pounds Professor of Management and professor of operations research, statistics, and operations management at the MIT Sloan School of Management, where she has been a faculty member since 1998. She investigates the theory and practice of analytics and its role in operations problems and is particularly interested in how to solve complex and practical problems in pricing, revenue management, supply chains, health care, transportation, and energy applications, among other areas. Since 2019, she has been the co-director of the Operations Research Center, an interdepartmental PhD program that jointly reports to MIT Sloan and the MIT Schwarzman College of Computing, a role in which she will remain. Perakis will also assume an associate dean role at MIT Sloan in recognition of her leadership.

Hare and Perakis succeed David Kaiser, the Germeshausen Professor of the History of Science and professor of physics, and Julie Shah, the H.N. Slater Professor of Aeronautics and Astronautics, who will be stepping down from their roles at the conclusion of their three-year term on Aug. 31.

“My deepest thanks to Dave and Julie for their tremendous leadership of SERC and contributions to the college as associate deans,” says Huttenlocher.

SERC impact

As the inaugural associate deans of SERC, Kaiser and Shah have been responsible for advancing a mission to incorporate humanist, social science, social responsibility, and civic perspectives into MIT’s teaching, research, and implementation of computing. In doing so, they have engaged dozens of faculty members and thousands of students from across MIT during these first three years of the initiative.

They have brought together people from a broad array of disciplines to collaborate on crafting original materials such as active learning projects, homework assignments, and in-class demonstrations. A collection of these materials was recently published and is now freely available to the world via MIT OpenCourseWare.

In February 2021, they launched the MIT Case Studies in Social and Ethical Responsibilities of Computing for undergraduate instruction across a range of classes and fields of study. The specially commissioned and peer-reviewed cases are based on original research and are brief by design. Three issues have been published to date and a fourth will be released later this summer. Kaiser will continue to oversee the successful new series as editor.

Last year, 60 undergraduates, graduate students, and postdocs joined a community of SERC Scholars to help advance SERC efforts in the college. The scholars participate in unique opportunities throughout, such as the summer Experiential Ethics program. A multidisciplinary team of graduate students last winter worked with the instructors and teaching assistants of class 6.036 (Introduction to Machine Learning), MIT’s largest machine learning course, to infuse weekly labs with material covering ethical computing, data and model bias, and fairness in machine learning through SERC.

Through efforts such as these, SERC has had a substantial impact at MIT and beyond. Over the course of their tenure, Kaiser and Shah have engaged about 80 faculty members, and more than 2,100 students took courses that included new SERC content in the last year alone. SERC’s reach extended well beyond engineering students, with about 500 exposed to SERC content through courses offered in the School of Humanities, Arts, and Social Sciences, the MIT Sloan School of Management, and the School of Architecture and Planning.

Read More

Covid-19 was somewhat of a metaverse itself. Many of our domains turned digital — with much attention toward one emerging space: virtual care. The pandemic exacerbated the difficulties of providing appropriate medical board oversight to ensure proper standard of services for patients. MIT researcher and former professor Amar Gupta explores through his research on how different states approach quality, safety, and coordination issues related to telemedicine and health care — and how we need to take an integrated approach to address the interoperability challenge and enhance care delivery.

Q: Since the onset of the global Covid-19 pandemic, how has the quality and landscape of patient care changed?

A: Covid-19 has served as a major catalyst for the adoption of virtual techniques in the U.S. and other countries around the globe. This adoption has occurred in many medical specialties, both in urban and rural areas. At the same time, it has raised several issues and challenges that need to be addressed on a priority basis.

In our recent research paper, we found that in the U.S., “the increased amount of virtual care during the Covid-19 pandemic has exacerbated the challenge of providing appropriate medical board oversight to ensure proper quality of care delivery and safety of patients. This is partly due to the conventional model of each state medical board holding responsibility for medical standards and oversight only within the jurisdiction of that state board and partly due to regulatory waivers and reduced enforcement of privacy policies.”

The prevailing restrictions, related to privacy of patient medical records and the ability for doctors from other states to see those records, were temporarily removed or made less prohibitive. This, in turn, can lead to situations where more medical images can go on an unauthorized basis into the public domain.

And then we have the overarching challenge of interoperability across medical practices and organizations, states, and countries. Years ago, it was just one doctor alone, or one medical system. Now a patient is going to multiple hospitals, multiple doctors. We find this creates issues with respect to treatment, as well as quality and safety of the patient, because the records are scattered or not easily accessed. Sometimes the same test is done two, three times over. Sometimes the records of another hospital are not looked at. Increasingly, medical professionals are complaining about the growing problem of information glut. Based partly on our previous work at successfully assisting major re-engineering and interoperability efforts in financial and defense industries, we believe that Covid-19 reinforced the urgent need for a broadly accepted global approach in the health-care interoperability arena.

Q: You recently published a paper about the impact of growing virtual care and the need for an integrated approach to enhance care delivery. Can you elaborate on your research study and subsequent proposal for the medical community?

A: The paper was started based on a presentation that I made in Washington, D.C., to a group of senior government officials about telemedicine, regulation, and quality control. The Federation of State Medical Boards then gave us names and addresses of the state medical boards in the U.S., and some abroad. We wrote to all of them with a questionnaire to find out what they were doing with respect to telemedicine.

A few of the questions we explored were: Do they have any standards for telemedicine in evaluating the quality of services being rendered? How do they deal with complaints? Have they received any complaints related to telemedicine?

We got responses from only some of the medical boards. What was clear is that there weren’t any uniform standards across the nation. In several states, there are two medical boards, one for allopathic medicine and one for osteopathic medicine.

It’s very difficult to be disbarred in the U.S. — the standards are very high. We found that there were cases when a doctor who had been disbarred from medical practice in one state was still practicing in another. There was also a case where the doctor had been disbarred in three states and was practicing in a fourth state.

We have instances of interstate telemedicine in the U.S., intercountry work in Europe, and intercontinental telemedicine today. Patients in the ICU at Emory University in Atlanta, for example, at nighttime, are seen by medical personnel working during day time in Australia. This is consistent with the model that we had proposed in our other paper to improve quality and safety of patients by addressing the consequences of circadian misalignment and sleep deprivation among doctors and other medical personnel.

We don’t want doctors who have been penalized in one city, state, or country going to another country and working there. Here, even within the country, this safeguard has not been historically true. For one, the Federation of the State Medical Boards itself has written that many people do not really register their complaints with them, which is cited in our research. There’s also a database available where state regulators can see what happened in other states with respect to specific doctors. That was used less than 100 times in 2017. In fact, two states used it for more than half of these cases. Some states never used it at all. They were basically neglecting what had happened to the doctor in other states, which was frightening.

The Federation of State Medical Boards recently developed a new technology to address this problem. They created an experimental website called docinfo.org, and they invited us to look at it. Using this site, we tried an experiment, by searching for a specific doctor who had been disbarred in three states. These database sites recommended that we have to go to the sites of the three state medical boards, and it actually took us there. When we got to the state medical boards, all the information has been redacted. This reminded me of write-only memory, where information is available somewhere, but nobody’s able to access it, which doesn’t really help the customer.

One of the state medical boards responded that “our state does not allow us to give any information under the Freedom of Information Act to anybody outside the state.” Another one, in our study, refused to give us any information, and said that, based on what we’ve written before, “I know what you’re going to do with this information. I’m not going to give it to you.”

The aspect of medical personnel other than doctors has been covered in a companion research paper: “Enhancing quality of healthcare and patient safety: oversight of physician assistants, nurses, and pharmacists in era of COVID-19 and beyond,” and its first reference asserts that medical error is the third major cause of death in the U.S.

People argue about the quality and cost of health care. If you look at the U.S. today, the cost per patient is the highest in the whole world. If you look at quality, the U.S. is generally ranked below all the other developed countries. In order to enhance quality and safety of health care as well as reduce overall cost, I propose that we need something like the equivalent of Jeanne Clery Act for health care, which “requires public and private colleges and universities to disclose information about certain crimes that occur on or near campus” — but related to doctors and other medical personnel.

If we have these types of techniques available, then patient-reported outcomes and the use of AI techniques will aid in getting our hands around how to improve health care not just for people, but for health care services and products, too. We really need to take that bigger initiative not only in this nation, but on a seamless basis around the world.

Q: With Covid-19, we saw the proliferation of AI-based solutions with predictive modeling, synthetic biology, and surveillance and contact monitoring. Predating the pandemic, robust AI models have enabled better forecasting, medical imaging, clinical workflows. What ongoing issues need to be addressed?

A: The definition of medicine has changed over the years. At one point, there was a doctor, and that doctor did most of the tasks. The nurse may be there, and a compounder to do the medications. The quality control issue was mainly on the doctor. Today, it’s a blend of the hospital network, doctors, bureaucrats, administrators. There are technical staff in charge of telemedicine systems and computer scientists who work on modeling.

Recently, I supervised a graduate thesis on prescription opioids, and we found that there was systematic discrimination. With white males, they were much more likely to be given the prescription. If it was a woman or a Black person, they were much less likely to get the pills, even with the same set of symptoms and issues. The graduate student also looked at the nurses records, and found that they were repeatedly saying, for one kind of patient, they were “less complaining,” and others were “complaining,” which in turn impacted the chance of getting the opioid prescription.

Now, trained AI models that assist in decision-making will also present bias. But in a situation like this, whom does one file a complaint against? Do you file it against the hospital? The doctor and nurse? The computer scientist?

In today’s world, as these systems are progressing from a single doctor to much more integrated system, it’s becoming more and more difficult to decide who is at fault. If they’re not taken care of earlier, we run the risk of large-scale harm.

AI-based networks are supposed to be trained and retrained at regular intervals using the latest data from a cohort of patients. As patients’ conditions change, and they take different drugs, the way they react to any other drug will be different. Few of these models are going through any retraining process.

About 15 years ago, I had coined the term “three-pronged approach” to describe my vision of evolving health care. The three-pronged approach means that there are people in proximity to the patient, maybe a nurse practitioner or family member who might be helping. There is a doctor who’s a domain expert who may be in another city, another state, another country. There’s IT and AI work that will take place.

The three-pronged approach to health care is very much in vogue today. To find effective solutions, we can’t look at a single prong — we need an integrated approach. While there are over 100 health-care interoperability efforts around the world which pertain to a particular geographic region or a particular medical specialty, we need to address the challenge of interoperability by devising and implementing a broadly accepted staged plan for global adoption, rather than just focusing at local, state, or national level. This, in turn, will also enable superior leveraging and management of health-care personnel, services, and products to support the global quest for health care for all: better, quicker, and less expensive.

Read More

Read More

It is a problem that has beguiled scientists for a century. But, buoyed by a $625,000 Distinguished Early Career Award from the U.S. Department of Energy (DoE), Matteo Bucci, an associate professor in the Department of Nuclear Science and Engineering (NSE), hopes to be close to an answer.

Tackling the boiling crisis

Whether you’re heating a pot of water for pasta or are designing nuclear reactors, one phenomenon — boiling — is vital for efficient execution of both processes.

“Boiling is a very effective heat transfer mechanism; it’s the way to remove large amounts of heat from the surface, which is why it is used in many high-power density applications,” Bucci says. An example use case: nuclear reactors.

To the layperson, boiling appears simple — bubbles form and burst, removing heat. But what if so many bubbles form and coalesce that they form a band of vapor that prevents further heat transfer? Such a problem is a known entity and is labeled the boiling crisis. It would lead to runaway heat, and a failure of fuel rods in nuclear reactors. So “understanding and determining under which conditions the boiling crisis is likely to happen is critical to designing more efficient and cost-competitive nuclear reactors,” Bucci says.

Early work on the boiling crisis dates back nearly a century ago, to 1926. And while much work has been done, “it is clear that we haven’t found an answer,” Bucci says. The boiling crisis remains a challenge because while models abound, the measurement of related phenomena to prove or disprove these models has been difficult. “[Boiling] is a process that happens on a very, very small length scale and over very, very short times,” Bucci says. “We are not able to observe it at the level of detail necessary to understand what really happens and validate hypotheses.”

But, over the past few years, Bucci and his team have been developing diagnostics that can measure the phenomena related to boiling and thereby provide much-needed answers to a classic problem. Diagnostics are anchored in infrared thermometry and a technique using visible light. “By combining these two techniques I think we’re going to be ready to answer standing questions related to heat transfer, we can make our way out of the rabbit hole,” Bucci says. The grant award from the U.S. DoE for Nuclear Energy Projects will aid in this and Bucci’s other research efforts.

An idyllic Italian childhood

Tackling difficult problems is not new territory for Bucci, who grew up in the small town of Città di Castello near Florence, Italy. Bucci’s mother was an elementary school teacher. His father used to have a machine shop, which helped develop Bucci’s scientific bent. “I liked LEGOs a lot when I was a kid. It was a passion,” he adds.

Despite Italy going through a severe pullback from nuclear engineering during his formative years, the subject fascinated Bucci. Job opportunities in the field were uncertain but Bucci decided to dig in. “If I have to do something for the rest of my life, it might as well be something I like,” he jokes. Bucci attended the University of Pisa for undergraduate and graduate studies in nuclear engineering.

His interest in heat transfer mechanisms took root during his doctoral studies, a research subject he pursued in Paris at the French Alternative Energies and Atomic Energy Commission (CEA). It was there that a colleague suggested work on the boiling water crisis. This time Bucci set his sights on NSE at MIT and reached out to Professor Jacopo Buongiorno to inquire about research at the institution. Bucci had to fundraise at CEA to conduct research at MIT. He arrived just a couple of days before the Boston Marathon bombing in 2013 with a round-trip ticket. But Bucci has stayed ever since, moving on to become a research scientist and then associate professor at NSE.

Bucci admits he struggled to adapt to the environment when he first arrived at MIT, but work and friendships with colleagues — he counts NSE’s Guanyu Su and Reza Azizian as among his best friends — helped conquer early worries.

The integration of artificial intelligence

In addition to diagnostics for boiling, Bucci and his team are working on ways of integrating artificial intelligence and experimental research. He is convinced that “the integration of advanced diagnostics, machine learning, and advanced modeling tools will blossom in a decade.”

Bucci’s team is developing an autonomous laboratory for boiling heat transfer experiments. Running on machine learning, the setup decides which experiments to run based on a learning objective the team assigns. “We formulate a question and the machine will answer by optimizing the kinds of experiments that are necessary to answer those questions,” Bucci says, “I honestly think this is the next frontier for boiling,” he adds.

“It’s when you climb a tree and you reach the top, that you realize that the horizon is much more vast and also more beautiful,” Bucci says of his zeal to pursue more research in the field.

Even as he seeks new heights, Bucci has not forgotten his origins. Commemorating Italy’s hosting of the World Cup in 1990, a series of posters showcasing a soccer field fitted into the Roman Colosseum occupies pride of place in his home and office. Created by Alberto Burri, the posters are of sentimental value: The (now deceased) Italian artist also hailed from Bucci’s hometown — Città di Castello.

Read More

Multivariable calculus, differential equations, linear algebra — topics that many MIT students can ace without breaking a sweat — have consistently stumped machine learning models. The best models have only been able to answer elementary or high school-level math questions, and they don’t always find the correct solutions.

Now, a multidisciplinary team of researchers from MIT and elsewhere, led by Iddo Drori, a lecturer in the MIT Department of Electrical Engineering and Computer Science (EECS), has used a neural network model to solve university-level math problems in a few seconds at a human level.

The model also automatically explains solutions and rapidly generates new problems in university math subjects. When the researchers showed these machine-generated questions to university students, the students were unable to tell whether the questions were generated by an algorithm or a human.

This work could be used to streamline content generation for courses, which could be especially useful in large residential courses and massive open online courses (MOOCs) that have thousands of students. The system could also be used as an automated tutor that shows students the steps involved in solving undergraduate math problems.

“We think this will improve higher education,” says Drori, the work’s lead author who is also an adjunct associate professor in the Department of Computer Science at Columbia University, and who will join the faculty at Boston University this summer. “It will help students improve, and it will help teachers create new content, and it could help increase the level of difficulty in some courses. It also allows us to build a graph of questions and courses, which helps us understand the relationship between courses and their pre-requisites, not just by historically contemplating them, but based on data.”

The work is a collaboration including students, researchers, and faculty at MIT, Columbia University, Harvard University, and the University of Waterloo. The senior author is Gilbert Strang, a professor of mathematics at MIT. The research appears this week in the Proceedings of the National Academy of Sciences.

A “eureka” moment

Drori and his students and colleagues have been working on this project for nearly two years. They were finding that models pretrained using text only could not do better than 8 percent accuracy on high school math problems, and those using graph neural networks could ace machine learning course questions but would take a week to train.

Then Drori had what he describes as a “eureka” moment: He decided to try taking questions from undergraduate math courses offered by MIT and one from Columbia University that had never been seen before by a model, turning them into programming tasks, and applying techniques known as program synthesis and few-shot learning. Turning a question into a programming task could be as simple as rewriting the question “find the distance between two points” as “write a program that finds the difference between two points,” or providing a few question-program pairs as examples.

Before feeding those programming tasks to a neural network, however, the researchers added a new step that enabled it to vastly outperform their previous attempts.

In the past, they and others who’ve approached this problem have used a neural network, such as GPT-3, that was pretrained on text only, meaning it was shown millions of examples of text to learn the patterns of natural language. This time, they used a neural network pretrained on text that was also “fine-tuned” on code. This network, called Codex, was produced by OpenAI. Fine-tuning is essentially another pretraining step that can improve the performance of a machine-learning model.

The pretrained model was shown millions of examples of code from online repositories. Because this model’s training data included millions of natural language words as well as millions of lines of code, it learns the relationships between pieces of text and pieces of code.

Many math problems can be solved using a computational graph or tree, but it is difficult to turn a problem written in text into this type of representation, Drori explains. Because this model has learned the relationships between text and code, however, it can turn a text question into code, given just a few question-code examples, and then run the code to answer the problem.

“When you just ask a question in text, it is hard for a machine-learning model to come up with an answer, even though the answer may be in the text,” he says. “This work fills in the that missing piece of using code and program synthesis.”

This work is the first to solve undergraduate math problems and moves the needle from 8 percent accuracy to over 80 percent, Drori adds.

Adding context

Turning math questions into programming tasks is not always simple, Drori says. Some problems require researchers to add context so the neural network can process the question correctly. A student would pick up this context while taking the course, but a neural network doesn’t have this background knowledge unless the researchers specify it.

For instance, they might need to clarify that the “network” in a question’s text refers to “neural networks” rather than “communications networks.” Or they might need to tell the model which programming package to use. They may also need to provide certain definitions; in a question about poker hands, they may need to tell the model that each deck contains 52 cards.

They automatically feed these programming tasks, with the included context and examples, to the pretrained and fine-tuned neural network, which outputs a program that usually produces the correct answer. It was correct for more than 80 percent of the questions.

The researchers also used their model to generate questions by giving the neural network a series of math problems on a topic and then asking it to create a new one.

“In some topics, it surprised us. For example, there were questions about quantum detection of horizontal and vertical lines, and it generated new questions about quantum detection of diagonal lines. So, it is not just generating new questions by replacing values and variables in the existing questions,” Drori says.

Human-generated vs. machine-generated questions

The researchers tested the machine-generated questions by showing them to university students. The researchers gave students 10 questions from each undergraduate math course in a random order; five were created by humans and five were machine-generated.

Students were unable to tell whether the machine-generated questions were produced by an algorithm or a human, and they gave human-generated and machine-generated questions similar marks for level of difficulty and appropriateness for the course.

Drori is quick to point out that this work is not intended to replace human professors.

“Automation is now at 80 percent, but automation will never be 100 percent accurate. Every time you solve something, someone will come up with a harder question. But this work opens the field for people to start solving harder and harder questions with machine learning. We think it will have a great impact on higher education,” he says.

The team is excited by the success of their approach, and have extended the work to handle math proofs, but there are some limitations they plan to tackle. Currently, the model isn’t able to answer questions with a visual component and cannot solve problems that are computationally intractable due to computational complexity.

In addition to overcoming these hurdles, they are working to scale the model up to hundreds of courses. With those hundreds of courses, they will generate more data that can enhance automation and provide insights into course design and curricula.

Read More