Photography credits: Photos of Pallavi Bharadwaj, Sarah Cen, Kimon Drakopoulos, Max Taylor, Mengdi Wang, and Heng Yang provided courtesy of the interviewees. Photo of Guy Bresler by Jennifer Donovan.
Massachusetts Institute of Technology Laboratory for Information and Decision Systems 77 Massachusetts Avenue, Room 32-D608
It is my privilege to introduce the latest issue This is
It is my privilege to introduce the latest issue of LIDS|All. This is my first introduction as LIDS director—a position I am excited and honored to hold, especially as a LIDS alum and faculty member. I would like to start by thanking the outgoing director, John Tsitsiklis, and outgoing associate director, Eytan Modiano. Their thoughtful leadership paved the way for new initiatives and expansion of the lab, and positioned LIDS with strength as a major research unit in MIT’s Schwarzman College of Computing. I know the shoes I am filling are big ones, and look forward to taking on the challenge.
Since its founding, LIDS has been one of the most prominent places where numerous fields have been transformed from a state of “gadgeteering” to one that proceeds along “well-established scientific lines,” as founding director Gordon Brown describes in his book Principles of Servomechanisms. This transformation has occurred through the development of rigorous analytical approaches together with their implementation in real-world applications through truly grandchallenge projects in the lab. Going forward, it is my hope that LIDS will continue to lead in this foundational research in the information and decision sciences, and its impactful applications in many domains, with the flavor that Gordon Brown and his colleagues practiced more than eighty years ago.
In the meantime, I am happy to share all of the terrific work happening at LIDS today. As we made our way through another period marked by the Covid-19 pandemic, you will see a different kind of year at LIDS reflected in the magazine — but a remarkable one nonetheless. Our com-
ABOUT LIDS
The Laboratory for Information and Decision Systems (LIDS) at MIT, established in 1940 as the Servomechanisms Laboratory, currently focuses on four main research areas: communication and networks, control and system theory, optimization, and statistical signal processing. These areas range from basic theoretical studies to a wide array of applications in the communication, computer, control, electronics, and aerospace industries.
munity continued to work and thrive remotely for much of the year, meeting the challenges of connecting from a distance with grace and creativity. While many traditions had to change direction (both the LIDS Student Conference and LIDS Graduation Celebration were held virtually, for instance) the core purpose of sharing our work and successes was kept beautifully whole, a testament to the spirit of the LIDS community in a time of much change.
I am deeply saddened, however, to report that our community also shared a tragic loss this past year. Graduate student Matthew Brennan, a rising star, generous researcher, and wonderful friend to many, died suddenly in January 2021 of a previously unknown medical condition. Supervised by Professor Guy Bresler, Matt explored theoretical aspects of machine learning and probability theory; at the 2020 Conference on Learning Theory (COLT), he won the Best Student Paper Award — one of many honors he had already received in a promising academic career. Matt brought this same boundless energy and passion to all of his endeavors, especially his friendships. He is sorely missed, and leaves behind a legacy of overwhelming kindness.
You can read more about Matt, who in many ways embodied the best of what it means to be part of LIDS, in this issue. You will also read about other exceptional people in our community including: graduate student Sarah Cen and her work in promoting trust and fairness in AI; faculty member Guy Bresler’s groundbreaking research in high-dimensional statistics; administrative assistant Max Taylor; alum Mengdi Wang’s work in reinforcement learning; postdoc Pallavi Bharadwaj’s vision for a renewable energy future; and graduate student Heng Yang and his research in robot perception.
Sincerely,
Sertac Karaman, Director
LIDS is truly an interdisciplinary lab, home to over 150 graduate students and post-doctoral associates from EECS, AeroAstro, and many other departments across MIT. The intellectual culture at LIDS encourages students, postdocs, and faculty to both develop the conceptual structures of the above system areas and apply these structures to important engineering problems.
To the Statistical Limit
By Greta Friar
When Guy Bresler, Associate Professor of Electrical Engineering and Computer Science at MIT and a member of LIDS, travels to attend conferences, he often spends extra time in the area checking out the local rock climbing. He likes to pick routes that are just on the edge of possible for his strength and abilities, where there’s uncertainty as to whether he can succeed. In his first few attempts of a hard climb, the individual moves can feel impossible, and
situation is that we’re totally in the dark, it’s not even clear how to think about these questions, what the right tools are, or what the right way to start is,” Guy says. “The risk is high, but personally, I find this style of research the most rewarding.”
One of the big questions that Guy wants to tackle is that of statistical-computational tradeoffs in statistical inference problems: what
“I like working on problems where the current situation is that we’re totally in the dark, it’s not even clear how to think about these questions, what the right tools are, or what the right way to start is.”
linking them together less probable still. Nevertheless, he keeps at it until he can (with luck) string together a sequence of moves that brings him to the top. One tough climb took him twenty tries, over ten separate occasions.
Guy’s attitude towards research is similar. He likes to tackle big questions for which the path to a solution is entirely unclear, with the only way forward being open-minded exploration.
“I like working on problems where the current
causes them to occur and how can we find the tipping points or phase transitions?
To understand his research, it helps to start with the basic question in mathematical statistics—how much data is needed to solve a problem? The minimum amount of data required to solve a problem is often called its statistical or information theoretic limit. Many problems for which this minimum amount of data has been determined are not actually practically solvable with this amount of data, because the
computational cost is too high. There is no computationally efficient algorithm that will work on them, so it might take a computer years to find the solution.
However, for many of these problems, if one gathered some further amount data, the very nature of the problem would shift and the computational cost to solve it would go down. This higher data threshold, sometimes referred to as the computational limit, is in laymen’s terms the amount of data needed to make the problem solvable on your laptop. Types of problems that have a mismatch between their statistically and computationally optimal data requirements are referred to as having a statistical-computational gap, and they include some of the most basic and universal statistical procedures, such as low rank matrix estimation, sparse phase retrieval, and robust sparse linear regression.
Finding such problems’ computational limits would provide valuable information to data scientists. If a scientist knows that collecting twice as much data will make solving a problem exponentially easier computationally, this would be crucial information that at a minimum might yield huge savings in cost and energy usage.
Guy hopes to find computational limits for as many problems as possible, but at a more basic level he wants to understand what changes about the nature of these problems as they
cross the computational limit to make them solvable by efficient algorithms. The shift is not simply a matter of scale, in which having more data speeds up the runtime for finding a solution using the same approach. Instead, the structural properties of a problem changes, enabling the use of completely different approaches—often quite simple ones—to solve it.
Even more than understanding individual problems, Guy is interested in finding new connections between seemingly different ones. Are there commonalities across problems that explain the qualitative shift they undergo at the computational limit? If there are commonalities, could discovering them allow researchers to more efficiently determine the computational limits and best algorithmic approaches for new problems?
To answer these questions, Guy wants to create a comprehensive average-case complexity theory for statistical inference problems. This will categorize statistical problems based on whether they are efficiently solvable or not, similar to the classical P versus NP theory for combinatorial problems. The challenge in relating statistical problems is that their inputs are data described by probability distributions, so connecting them requires also precisely relating these distributions, a delicate task. Demonstrating equivalence between complex problems is difficult, but it has the potential for huge payoff: it will allow researchers to extend knowledge about one problem’s statistical and
computational limits to others, and transfer tools and approaches from better-understood problems to less-understood ones.
Building a comprehensive theory of this scale is a huge undertaking, and as such requires collaboration. Guy’s closest collaborator on this project had been his graduate student Matthew Brennan, but Matt passed away in early 2021. Guy still grapples with his loss as a friend, rising star in the field, and partner in the work, and for a while he had to set the project aside. However, in recent months he has turned his attention to it again. In the Fall of 2021 he lead a program, “Computational Complexity of Statistical Inference,” at the Simons Institute at the University of California Berkeley. The program hosted researchers from a variety of adjacent fields to work on the topic.
“We learn the most and have the most fun when we work as a team,” Guy says, explaining one of the keys pieces of his approach: Whether in climbing or in research, when you take on a challenge that tests your limits, the best way to succeed is to tackle it together.
HONING ROBOT PERCEPTION
By Grace Chua
Against the odds, LIDS PhD student Heng Yang (who goes by Hank) has had a rather good year.
For graduate students who spent much of 2020 working and attending classes remotely, life during the pandemic’s early months was a hectic, monotonous hum of research. When you’re immersed in your work, “seven days a week look alike, and this is especially the case during the pandemic,” Hank said.
“I
WANTED TO SEE DIFFERENT THINGS. I WANTED TO EXPLORE.”
To help break up the monotony, Hank started 2021 by co-organizing the 26th annual LIDS Student Conference, which was hosted virtually for the first time, so that even during the pandemic the LIDS community could stay connected and inspired. Amid student talks on control, machine learning methods, and other topics, the conference team also organized a musical performance to mark the closing and help students feel a sense of belonging from afar. Not long after, an algorithm Hank developed— together with his supervisor, Professor Luca Carlone, Vasileios Tzoumas (a LIDS post doc at the time), and LIDS graduate student Pasquale
Antonante—was integrated into a MATLAB suite of navigation tools that companies use for commercial and industrial robotics systems.
Then, Hank presented his work at international conferences on robotics and computer vision. He honed his communication skills with a three-minute MIT Research Slam video. He has several papers in press and is due to complete his PhD at the end of the 2021-2022 school year. And in the coming months he will hit the job-talk circuit.
How did the earnest, energetic Hank get here? He quips that his career so far has been a ‘random walk’.
Hank grew up in China’s Jiangsu Province, just north of Shanghai. At the prestigious Tsinghua University he graduated with top honors, exploring the seemingly disparate topics of automotive engineering and the mechanics of how honeybees drink.
When he came to MIT for his graduate work, he brought this same spirit of intellectual curiosity, but was unclear about what direction to follow. “Five or ten years ago, I didn’t know what I wanted to do. I wanted to see different things. I wanted to explore,” he explains. And while he was able to do this with much success at MIT, he also feels “there was some luck or fortune involved in finding this path.”
It was while researching medical imaging
methods for his master’s degree that Hank took a course with Professor Russ Tedrake and became interested in how to combine theory with practical robotics applications. That led him to LIDS, where theory and practice rub shoulders daily.
At LIDS, Hank has very happily found a home in SPARKLab, the research group led by his supervisor, Professor Luca Carlone. Here, Hank and his colleagues work on algorithms for robot perception, among other projects.
For robots like self-driving cars, perception means sensing the external world and using hardware and algorithms to create an internal model of the world around it.
For a basic example, a self-driving car would have to take a snapshot of an oncoming car. Then, it feeds this image through a neural network that detects keypoints in the 2-dimensional image – door handles, wheels, headlights – and matches them to its prior information about what cars look like in 3D.
That allows the system to figure out, and more importantly track, whether the oncoming car is hurtling toward you, or simply passing by. And it has to do all this in a split second.
But current keypoint detection methods can produce a lot of outliers, like mistakenly identifying part of a tree as part of a car. So how can you make the system robust against a large percentage of these outliers? How does the robot know that its output - ‘Oncoming Toyota sedan at 50 yards’ - is correct?
Together with colleagues, Hank developed the graduated non-convexity (GNC) algorithm, which finds a single best solution for matching the 2D image with the 3D model, then keeps checking and refining it until an optimal solution is found.
He is also designing algorithms that can be certified to be right under certain conditions. Used in tandem with the GNC algorithm, this produces an extra level of certainty when dealing with noisy, complex images such as moving traffic: you can say that your solution is not only
optimal but also safe. In future, perhaps, these algorithms will become part of the way neural networks are trained to sense and identify keypoints.
So what’s next for Hank?
In the next few years, he says, he hopes to work on the problem of active perception.
To get an understanding of active perception, you can start by imagining a robot arm that needs to grab an object that’s only partially visible.
Humans do this automatically: if we’re reaching for a document under a pile of books, or a mug amid glasses in a cabinet, we move our hands and bodies to lift the books or move the glasses and pick up the item. We use our actions to help us perceive better.
A robotic system would need to use its sensors and algorithms to identify and track the object, and move itself to reach out and pick it up. If the object is moving, things get even more complicated.
Vasileios Tzoumas, now an assistant professor at the University of Michigan, says: “Believe it or not, there is almost no work on this domain, so it’s the right time for researchers to start on this problem.”
tions, says Vasileios. “He has the capacity to start working on a hard problem and, within a year or two, produce a robust concrete solution with almost no weaknesses.” Most importantly, he adds, Hank is always generous and willing to help others.
Hank may joke his career has been a random walk. But random walks reveal many real underlying patterns – such as the trajectory of a thoughtful and curious rising star.
Hank seeks out problems that have real-world applications, and keeps improving on his solu-
USING DATA SCIENCE FOR GOOD IN COVID-19
By Greta Friar
LIDS alum Kimon Drakopoulos SM ’11, PhD ̓16 focused his doctoral research on the analytics of contagion and epidemics. When the Covid-19 pandemic hit, Kimon, now an assistant professor of Data sciences and Operations at the University of Southern California’s (USC) Marshall School of Business, felt an imperative to put his skills for analyzing epidemics to good use.
“We have been working on all of this cool data science and making these complex models, but if we don’t actually apply those analytics towards mitigation of the disease, then we are wasting these resources,” Kimon says.
Kimon, looking for an opportunity to be helpful, took note when his home country of Greece announced that it was planning to reopen its borders to travel after a strict lockdown early in the pandemic. At the time, resources for managing Covid-19 were limited. Greece, like other countries, would not be able to test every traveler for Covid-19 before letting them enter the country. Instead, policymakers would have to devise a strategy for both testing and enacting restrictions selectively in order to minimize spread of the disease without depressing travel—and its economic benefits—more than necessary. Most countries dealing with this problem were using a combination of random testing and travel restrictions based on countrywide publicly reported epidemiological statistics: travelers coming from a country with very
high case numbers, or deaths, would be more likely to be tested or temporarily banned from entering the country. However, Kimon knew that that strategy would miss a lot of cases. AI could help.
“Algorithms are particularly useful for decision making in cases like this when there are complicated dynamics, limited resources, and high uncertainty,” Kimon says.
Confident that he could contribute, Kimon emailed Greek Prime Minister Kyriakos Mitsotakis to offer his assistance. Within hours, the PM had accepted his offer to meet. Soon, Kimon would find himself running major aspects of Greece’s Covid-19 border management strategy. For months, he did little else but work on this one problem. Kimon took on a variety of duties, such as managing data science and engineering teams, coordinating with other groups involved in the border screening system, and training personnel who interact with travelers. He and his coauthors Hamsa Bastani (Wharton) and Vishal Gupta (USC) spent the peak travel months of 2020 getting only two hours of sleep a night.
First, Kimon and collaborators designed Eva, a reinforcement learning system rolled out in summer 2020, that recommended which travelers to test in order to catch asymptomatic cases. Travelers coming to Greece would fill out a form with basic demographic information.
Based on this, Eva would recommend whether they should be tested. Tested travelers would quarantine for up to 48 hours until they received their results, and those results would ultimately be fed back to Eva to improve estimates and drive the next round of recommendations.
as many cases as possible from travelers whose demographics were currently deemed high risk; and exploration, testing other travelers to identify demographics that might become high risk in the near future. With a virus like Covid-19 in which cases can surge very quickly, this exploratory testing is key to an effective strategy, and is part of what’s missing in a strategy based solely on countries’ current epidemiological statistics.
Implementing Eva required careful coordination in order to protect the privacy of travelers’ health information. Kimon worked closely with government officials and lawyers to set up a system that met the requirements of the European Union’s General Data Protection Regulation (GDPR). Eva could not be fed live test results, but instead received batches of pseudonymized data that included only the demographic information deemed necessary for the algorithm to make good predictions. Kimon and his team learned to work around this sort of limitation.
“The exciting part about doing something applied, when you are thinking about real world constraints, is that all the stylized math that you already know just collapses and you have to be creative. You need to know the fundamentals, so that you can adapt them quickly and effectively,” Kimon says.
The algorithm made recommendations with two goals in mind: exploitation, trying to catch
Their solutions proved effective. The researchers calculate that Eva was as much as four times
better than random testing at catching asymptomatic cases during peak travel times, and up to 45% more effective than testing based on epidemiological statistics.
Optimizing Eva was only part of the researchers’ challenge. Kimon’s team also met with the COVID-19 Executive Committee of Greece twice a week to inform policymakers’ decisions on travel restrictions and testing requirements. Kimon worried at first about how well the data scientists and policymakers would be able to communicate, but he found their exchanges rewarding and came to appreciate the difficulty of balancing the data his team provided against various other factors when deciding on policies. He says that the key to good collaboration was transparency. Data scientists might like to build complex models, but when providing information to a policymaker, Kimon thinks it’s important to give them something unambiguous and actionable.
“Simple, deep insights make the world move,” Kimon says.
Eventually, as testing resources became less scarce and vaccines became available, there was no longer need for Eva’s recommendations. Over the past year, Kimon has returned to his normal life, teaching, conducting research, and getting a full night’s sleep. His current projects focus on a different sort of viral spread, that of misinformation on social networks. He remains
passionate about the use of data science and analytics for governance. And after spending so long managing other people’s travels, Kimon managed to go on a trip of his own. He and his wife (fellow LIDS alum Yola Katsargyri SM’08, PhD ’17) spent a month in Alaska, an experience he describes as incredible. Kimon does not miss the exhaustion, media attention, or disruption to his life that came with his time working on Covid-19, but he remembers the experience with gratitude.
“The opportunity to work on a high impact project like this is rare, perhaps once in a lifetime, and I feel blessed that I had the chance,” Kimon says. “I am extremely proud of what my team accomplished.”
MOVING BEYOND GAMES
By Greta Friar
When Mengdi Wang SM ’10 PhD ’13 was a graduate student at LIDS she took a course on poker. Mengdi played often at MIT, honing her strategy and soon she was winning cash prizes in large poker tournaments. Now an associate professor of electrical and computer engineering at Princeton University, Mengdi does not play much poker these days. But when she recalls her training at LIDS, those poker games remain a favorite memory.
Reinforcement learning, the field of machine learning in which Mengdi works, also spent much of its early years focused on how to win games. Now, researchers including Mengdi are working to improve the methodology behind reinforcement learning in order to make it scalable, efficient, and generalizable, so that its use can be expanded in fields from medicine to finance to robotics.
In reinforcement learning, the AI agent is presented with an environment in a certain state— say, a game of poker where it holds a pair of aces—and must choose what action to take—in this case check, call, raise, or fold. The AI receives a positive or negative reward for its actions, and its goal is to maximize its cumulative reward. In this way, the AI learns through trial and error, not unlike a human child. The AI develops its own rules for how to maximize rewards. Once trained, algorithms are capable of stringing together an optimal sequence of decisions in a very complex environment, developing strategies that surpass those of humans. AIs trained with reinforcement learning
have bested human champions in games from chess to Go to Starcraft. A single algorithm, DeepMind’s MuZero, used reinforcement learning to master sixty different games without being given any of the rules at the start.
The strengths that enable these algorithms to best human players also make them alluring potential tools in other fields. “That gives researchers the hope that computers could surpass humans in many other domains, and help us make better decisions,” Mengdi says.
The potential uses for reinforcement learning are manifold. It has been used to varying success in smart device networks, stock trading algorithms, natural language processing, efficient energy systems, self-driving cars, robotics, and more.
However, there are challenges to expanding the use of reinforcement learning. When Mengdi began exploring this area of interest, she found that there were many open theoretical questions in the field. She describes her work as establishing the foundational methodology for reinforcement learning algorithms, in order to help unlock the full potential of these algorithms. Although early in her career, Mengdi has already made significant contributions to reinforcement learning methodology and theory, as well as developed algorithms to solve a variety of practical problems.
One of Mengdi’s key interests is efficiency: are algorithms making the best use of data and ex-
periments? Training tends to require an algorithm to search large and complex problem spaces, which consumes a lot of energy and runtime. Mengdi’s group has developed methods that provably enable algorithms to efficiently determine the optimal strategy, or policy, to follow.
Another of Mengdi’s interests is moving beyond the use of additive rewards, which have been the cornerstone of reinforcement learning, but
have limitations. Game strategies are often easy to deconstruct into discrete moves that can be assigned values. However, some complex problems cannot be so easily deconstructed. Furthermore, researchers may want the AI to incorporate other factors, such as exploration, risk, and prior human knowledge, into its optimal policy. Mengdi has made progress in this area as well: “We show that with not too complex modifications to existing algorithms, one can actually extend the entire problem domain to solve problems with complex utilities that are not additive,” Mengdi says.
Mengdi has also tackled how to train an AI with small or sparse datasets. Gaming algorithms can be trained in a simulator that allows them to run through many, many iterations of the game and so to encounter most of the possible states they will ultimately be faced with as they develop their optimal strategy. In fields such as medicine, the problem space tends not to be so well mapped out, and so the algorithm must develop a strategy based on limited information. Mengdi has experienced this issue first hand. When she worked with medical insurance companies to optimize the treatment plan for people receiving knee replacement surgery, the only data available were several hundred clinical claims, a very small dataset from an AI perspective. Mengdi’s algorithm was still able to improve on previous strategies, but not by as much as she thinks it could have with better data. Mengdi’s group is working on methodology to help improve algorithms’ success when faced with such data limitations.
“I think the real impact will be when we make reinforcement learning work on real systems that cannot be simulated like computer games, so that we can solve practical problems much better,” Mengdi says.
As Mengdi works on advancing the methodology in her field, she also celebrates advances in the inclusivity of the field. When she was starting out, there were few women faculty members whose example she could follow. Female students getting into the field now have many examples—Mengdi included—to show them that success is possible.
“In academics there is this problem of gender bias and gender imbalance. Machine learning, perhaps because it is a newer field, I think is more diverse, and it’s been exciting to see that diversity growing in recent years,” Mengdi says. “I work with amazingly good female students and researchers, and I cannot wait to see what the younger generation achieves.”
For young researchers and future algorithms alike, Mengdi believes that the sky is the limit.
A RENEWABLE ENERGY FUTURE
By Grace Chua
As clouds of smoke billowed from the power plant smokestack, Pallavi Bharadwaj thought, This isn’t sustainable.
Each day, the electrical engineering student passed a coal-fired power plant on her way to university from her family’s home in Delhi. Between car exhaust and fossil fuels, the city was densely blanketed in some of the world’s worst smog. Pallavi resolved to find a renewable energy solution.
the Indian government for a six-month program, and took a class with LIDS Senior Research Scientist Marija Ilic to learn more about the theory and modeling of complex power systems.
The class introduced her to interdisciplinary research at LIDS and since January 2020, she has been a postdoctoral researcher at the lab, working with Marija on the next generation of electric energy systems.
“As electrical engineers, most of us are schooled in one specific way of thinking,” Pallavi says. “At LIDS, there are such brilliant minds working on the data side of things; you should take advantage of all these different perspectives.”
Her quest led her to the leafy Bangalore campus of the Indian Institute of Science, where she spent her PhD developing hardware and algorithms for solar photovoltaic systems to continually and more efficiently convert sunlight to electricity. The technology Pallavi developed cost a fraction of commercial systems’ price and received a patent in October 2021.
But the compact, high-energy Pallavi, whose first name means ‘new leaf’ or ‘green shoots’ in Hindi, hasn’t rested on her laurels.
In fall 2019, she arrived in Boston as a PhD exchange student, one of a handful selected by
“As electrical engineers, most of us are schooled in one specific way of thinking,” Pallavi says. “At LIDS, there are such brilliant minds working on the data side of things; you should take advantage of all these different perspectives.”
In her time at the lab, together with a team of fellow postdocs and students, Pallavi developed a mathematical model of a large airport’s power system to help improve its energy efficiency.
Large industrial systems, such as a large airport’s, often involve many different sources and types of energy: they may draw electricity from the grid, but also generate their own electric-
ity from solar panels, use gas for heating water, and so forth—making it a complex multi energy system.
Pallavi and her colleagues first developed a common language of modeling variables to make these different energy systems ‘speak to each other’ so they can be integrated. This work is a part of fundamental energy-based modelling research Marija’s group introduced recently. Then, they demonstrated that their methods worked in practice in a large industrial HVAC (heating ventilation and air conditioning) system to make it energy and cost efficient.
In the last few months, Pallavi has enjoyed interacting with several national labs and academic institutes across US to launch another challenging project to boost the cybersecurity of microgrids. With recent events across the globe where cyber-attacks threatened the normal operation of electrical and transportation systems, cyber security is emerging as the need of the hour with exponential growth in energy electrification.
When a cyberattack occurs, hackers may mask the attack by altering sensor readouts to say all is well. So Pallavi and Dan Wu (a fellow postdoc) created an observer for microgrid—a set of algorithms that can give an estimate of the internal state of a system based on its inputs and outputs—that learns to recognize the grid’s normal state, and course-corrects when it spotssomething amiss. By creating their observer,
which predicts correct system behavior even under compromised sensor measurements, Pallavi and her colleagues have helped move toward more secure microgrids, and more stable energy supplies for those using them.
In between her research projects, Pallavi has found time to organize wellness activities for MIT’s postdoctoral researcher community during the pandemic, from sunset yoga to a guided meditation session. “Postdocs are often balancing work and family, and in a transition phase of their lives,” she says. “It can be a very stressful, isolating experience, so connecting with others can really heal you.”
Marija describes Pallavi as a motivated, organized self-starter: “I’m positive she will work well with potential partners and sponsors and become a successful faculty member.”
That’s exactly where Pallavi is headed next. She is currently in the decision-making phase with several faculty and industrial position offers she has received. She hopes to expand on her work with power engineering, specifically focusing on the transition to renewable energy.
Another question Pallavi hopes to explore is how to optimize energy storage solutions. Interestingly, the growing popularity of electric vehicles offers a possible part of the solution, for what is a plug-in electric vehicle but a rechargeable battery on wheels? The challenge of developing battery management systems for not only increased lifetime of batteries but also
recharge them optimally with renewable energy sources will bring forth the real green energy transition.
But there are a few snags. Both electric vehicles and solar cells use direct current, and power grids use alternating current, which is easily converted to higher or lower voltages for ease of transmission or safety. And while solar panels generate electricity during the day, most electric vehicles are parked and plugged in at night.
A smarter grid would work out, through algorithms and pricing incentives, how and when to connect and integrate all these different components to generate and store electricity from renewables most efficiently.
As Pallavi sees it, she could readily work on this for much of her future career. “Big policies are being made for climate change mitigation. But who will make the energy transition happen? Our job in the next 30 years is to make the net-zero transition technically possible.”
LIVING BETTER WITH ALGORITHMS
by Grace Chua
LIDS PhD student Sarah Cen remembers the lecture that sent her down the track to an upstream question.
At a talk on ethical AI, the speaker brought up a variation on the famous trolley problem, describing the following scenario: Say a selfdriving car is traveling down a narrow alley with an elderly woman on one side, a small child on the other, and no way to thread between both without a fatality. Who should the car hit?
Then the speaker said: Let’s take a step back. Is this the question we should even be asking?
That’s when things clicked for Sarah. Instead of considering the point of impact, a self-driving car could avoid choosing between two bad outcomes by making a decision earlier on. The speaker pointed out that the car should have determined that the space was narrow when entering the alley and slowed to a speed that kept everyone safe.
Recognizing that today’s AI safety approaches often resemble the trolley problem, focusing on downstream regulation such as liability after someone is left with no good choices, Sarah wondered: What if we could design better upstream and downstream safeguards to such problems? This question has informed much of Sarah’s work.
the social systems on which they intervene,” Sarah says. Ignoring this fact risks creating tools that fail to be useful when deployed or, more worryingly, that are harmful.
Sarah arrived at LIDS in 2018 via a slightly roundabout route. She first got a taste for research during her undergraduate degree at Princeton University, where she majored in mechanical engineering. For her master’s degree, she changed courses, working on radar solutions in mobile robotics (primarily, for self-driving cars) at the University of Oxford. There, she developed an interest in AI algorithms, curious about when and why they misbehave. So, she came to MIT and LIDS for her doctoral research, working with Professor Devavrat Shah in the Department of Electrical Engineering and Computer Science, for a stronger theoretical grounding in information systems.
Together with Devavrat and other collaborators, Sarah has worked on a wide range of projects in her time at LIDS. One such project focuses on a method for translating human-readable social media regulations into concrete auditing procedures.
“Engineering systems are not divorced from
Suppose, for example, that regulators require that any social media content containing public health information not be vastly different for left- and right-leaning users. How should auditors check that a social media platform complies with this regulation?
Designing an auditing procedure is difficult in large part because there are so many stakeholders when it comes to social media. Auditors have to inspect the algorithm without accessing sensitive user data. They also have to work around tricky trade secrets, which can prevent them from getting a close look at the very algorithm that they are auditing because these algorithms are legally protected. Other considerations come into play as well, including balancing the removal of misinformation with the protection of free speech.
To meet these challenges Sarah and Devavrat developed an auditing procedure that does not need more than black-box access to the social media algorithm (which respects trade secrets), does not remove content (which avoid issues of censorship), and does not require access to users (which preserves users’ privacy).
In their design process, the team also analyzed the properties of their auditing procedure, finding that it ensures a desirable property they call decision robustness. As good news for the platform, they show that a platform can pass the audit without sacrificing profits. Interestingly, they also found the audit naturally incentivizes content diversity, which is known to help reduce the spread of misinformation, counteract echo chambers, and more.
In another line of work, Sarah looks at whether people can receive good long-term outcomes when they compete for resources without knowing upfront what resources are best for them.
Take, for example, the process of finding employment. Workers want to be matched with employers, and vice versa. In this matching market, both workers and employers have matching preferences: workers prefer some jobs over others, and employers prefer some qualifications over others. However, workers and employers need to learn these preferences. For instance, workers may learn their job preferences from internships.
But learning can be disrupted by competition. If workers with a particular background are repeatedly denied jobs in tech due to high competition, for instance, they may never get the knowledge they need to make an informed decision about whether they want to work in tech. Similarly, tech employers may never see
and learn what these workers could do if they were hired.
Sarah and Devavrat’s work examines the interaction between learning and competition, studying whether it is possible for individuals on both sides of the matching market to walk away happy. They focused on four criteria: stability, low regret, fairness, and high social welfare. Interestingly, they found that it is indeed possible to achieve all four simultaneously and discussed what conditions make this outcome possible.
For the next few years Sarah plans to work on a new project, studying how to quantify the effect of an action X on an outcome Y when it’s expensive—or impossible—to measure this effect directly and focusing on systems with complex social behaviors.
For instance, during the height of the COVID-19 pandemic, many cities had to decide what restrictions to adopt, such as mask mandates, business closures, or stay-home orders. They had to act fast and balance public health with community and business needs, public spending, and a host of other considerations.
Typically, in order to estimate the effect of each restriction on infection rates, one might compare the infection rates in areas that adopted different restrictions. If one county has a mask mandate while its neighboring county does
not, one might think comparing the counties’ infection rates would reveal the effectiveness of mask mandates.
But of course, no county exists in a vacuum. If, for instance, people from both counties gather to watch a football game in the maskless county every week, these counties mix. These complex interactions matter, and Sarah plans to study questions of cause-and-effect in such settings.
“We’re interested in how decisions or interventions affect an outcome of interest, such as how criminal justice reform affects incarceration rates or how an ad campaign might change the public’s behaviors,” Sarah says.
Sarah has also stayed involved in the MIT community. As one of three co-presidents of the Graduate Women in MIT EECS student group, she helped organize the inaugural GW6 research summit featuring the research of women graduate students – not only to showcase positive role models to students, but also to highlight the many successful graduate women at MIT.
Whether in computing or in the community, a system taking steps to address bias is one that enjoys legitimacy and trust, Sarah says. “Accountability, legitimacy, trust—these principles play crucial roles in society and, ultimately, will determine which systems endure with time.”
Sound Bites: Max Taylor
What do you do at LIDS?
I support Professors Berwick, Bresler, Ozdaglar, Shah, Tsitsiklis, and Wilson. The way I describe what I do at LIDS is I do all of the administrative things that help the faculty do their research. So I coordinate meetings and manage inboxes and get people reimbursed and
sort of all of the “other tasks as assigned” that help the faculty and the grad students do their work. I’m also learning how to do some LIDS-wide work with our databases and account reporting. Where did you work before coming to LIDS
and MIT?
I worked at Bard Academy at Bard College at Simon’s Rock. I was a residence director there for two and a half years, working with 9th and 10th grade students—lived in the building with them, shared my dog, Rocky, with them, the whole thing. I was basically their parent in place, helping them get used to being in school away from their families. It was intense, but also a really unique and interesting way to live that didn’t necessarily feel like work all the time.
What brought you to MIT?
Working with those younger kids in the boarding school setting was a big a dream for a while—I actually moved
from L.A. to New England to do that. My partner came with me. He was also a residence director at Simon’s Rock. But what he always wanted to do was to focus more on DEI and identity resource work. So when a DEI program administrator position opened up at Tufts, he applied and got it, which was really exciting! I came with him to Boston and started looking for jobs in higher ed. A really close friend of mine did his bachelor’s and master’s at MIT, and I kept hearing about all the cool things that happen here, so I kept applying for positions until something came through.
What do you do outside of work?
I’m finishing up my bachelor’s of science in psychology at Northeastern, as a foundation to working in education and figuring out how to best work with students. My career in student services started very briefly at a therapeutic school—so working with students who had been to rehab or had been to wilderness therapy—and I realized that if the people that worked with them had a better understanding of what they were going through maybe
things wouldn’t have been as hard. So I took that to heart when I chose my degree.
But when I’m not doing course work— I just have my capstone project left—I hang out with my dog and my partner. We’re still exploring Boston since we’ve only been here a few months. We’re really big on exploring for food, especially. We have our favorite Italian restaurant in the North End already, and keep looking for where the best of our favorite foods can be found.
In honor of our 2021 graduates, LIDS once again hosted a virtual commencement reception held as part of a larger IDSS-wide celebration.
This special celebration featured remarks from faculty and graduates, and for the LIDS portion, a unique opportunity to highlight each graduate individually. While we look forward to the return of in-person festivities, it also remained a wonderful feature of the virtual format that family and friends from around the world, who might not otherwise have been able to attend, were able to join the celebration.
Our congratulations to the amazing class of 2021!
PhDs received by:
Raj Agrawal
Anastasiya Belyaeva
Phil Chodrow
Sagar Indurkhya
Rupamathi Jaddivada
Igor Kadota
Oscar Mickelin
David Miculescu
Dennis Shen
Dogyoon Song
Rajat Talak
Paxton Turner
Zhi Xu
Jinglong Zhao
SMs or MEngs received by:
Janak Agrawal MEng
Kwang Jun Ahn SM
Abdullah Alomar SM
Yun Chang SM
Romain Cosson MEng
George Denove SM
Georgia Dimaki SM
Joshua Fishman SM
Sarah Flanagan MEng
Sule Kahraman MEng
Lukas Lao Beyer MEng
Haochuan Li SM
Eshaan Nichani MEng
Neha Prasad MEng
Saeyoung Rho SM
Abhin Shah SM
Jingnan Shi SM
Yi Tian SM
Samuel Ubellacker MEng
Héctor Javier Vazquez MEng
William Wei Wang SM
Xinyu Wu SM
Cindy Yang MEng
Alexandra Zytek SM
LIDS Awards & Honors
Awards
Professor Dimitri Bertsekas received the 2022 IEEE Control Systems Award.
Student Matthew Brennan (supervised by Professor Guy Bresler) won the Best Student Paper Award at COLT2020. Notably, he also received this award at COLT2018, for related work, with Professor Bresler and collaborator.
Professor Tamara Broderick was awarded an ONR Early Career Grant.
Professor Luca Carlone was a Best Conference Paper Finalist at the International Conference “Robotics: Science and Systems” (RSS 2021). He also received the Outstanding Associate Editor Award at ICRA 2021 and the distinction of Outstanding Reviewer at CVPR 2021. Further honors include an NSF CAREER Award, 2020 Best Paper Award Honorable Mention from the IEEE Robotics and Automation Letters, and Track Best Paper Award at the 2021 IEEE Aerospace Conference.
Professor Jonathan How received a 2020 Best Paper Award Honorable Mention from the IEEE Robotics and Automation Letters.
Staff member Francisco Jaimes received the Schwarzman College of Computing Infinite Mile Award for Diversity & Community.
Student Igor Kadota received the MIT School of Engineering Graduate Student Extraordinary Teaching and Mentoring Award.
Student Dennis Shen, supervised by Professor Devavrat Shah, won the George M. Sprowls PhD Thesis Award in Artificial Intelligence and Decision Making.
Professor David Simchi-Levi was awarded the INFORMS Impact Prize.
Honors
Professor Tamara Broderick was awarded membership in the 2021 COPSS Leadership Academy.
Professor Luca Carlone was promoted to Associate Professor Without Tenure by the Department of Aeronautics and Astronautics, effective July 1, 2021.
Professor Jonathan How was elected to the National Academy of Engineering in 2021.
Senior Research Scientist Marija Ilic was elected to the National Academy of Engineering in 2021.
Professor Asu Ozdaglar was named a 2021 IEEE Fellow.
Professor Alexander Rakhlin was promoted to full professor by the Department of Brain and Cognitive Sciences, effective July 1, 2021.
Professor Devavrat Shah was named the Andrew (1956) and Erna Viterbi Professor of Electrical Engineering and Computer Science by the EECS department, effective July 1, 2021.
Professor Caroline Uhler was made the co-director of the newly-launched Eric and Wendy Schmidt Center, a center at the Broad Institute devoted to understanding the programs of life. Professor Uhler was also elected a Council Member of the International Statistical Institute.
LIDS COMMUNITY
The 2020-2021 academic year saw the LIDS community adapting to a largely virtual experience as MIT and the world navigated the Covid-19 pandemic.
While this pivot created both new challenges and opportunities, the LIDS community found ways to stay connected and engaged, transforming events into successful remote experiences.
Our students and postdocs continued to play a key role in organizing these activities. Through their work on different committees, they organized weekly virtual social events and weekly virtual LIDS & Stats Tea Talks (a popular series of informal research presentations); as well as a a virtual session on faculty position applications featuring both current LIDS faculty and LIDS alums.
Our thanks to all of the students, faculty, and staff who made these a success! We’d like to thank here, in particular, the student and postdoc organizers:
LIDS Social Committee
Chirag Rao
Soumya Sudhakar
Jerrod Wigmore
LIDS & Stats Tea Talks Committee
Kwangjun Ahn
Xinzhe “Roger” Fu
Brice Huang
Chandler Squires
Heng (Hank) Yeng
LIDS Postdoc Committee
Mehdi Jafari
Bharadwaj Satchidanandan
Silun Zhang
LIDS Seminars 2020-2021
Seminars are a highlight of the LIDS experience. Each talk, which features a visiting or internal invited speaker, provides the LIDS community an unparalleled opportunity to meet with and learn from scholars at the forefront of their fields.
Listed in order of appearance.
Francis Bach
INRIA
Computer Science Department at Ecole Normale Supérieure
Mor Harchol-Balter
Carnegie Mellon University
Department of Computer Science
Michel Delfour
Université de Montréal
Department of Mathematics and Statistics
Mary Wootters
Stanford University
Computer Science Department
Department of Electrical Engineering
Andreas Krause
ETH Zürich
Computer Science Department
Magnus Egerstedt
Georgia Institute of Technology
School of Electrical and Computer Engineering
Kamalika Chaudhuri
University of California, San Diego
Department of Computer Science and Engineering
Daniel Spielman
Yale
Department of Mathematics
Department of Computer Science
David Simchi-Levi
MIT
Laboratory for Information and Decision Systems
Institute for Data, Systems, and Society
Department of Civil and Environmental Engineering
Todd Coleman
University of California, San Diego
Department of Bioengineering
John Langford
Microsoft Research
Research Groups: Machine Learning, AI for Systems, and Reinforcement Learning
Jeff Shamma
University of Illinois at Urbana-Champaign
Industrial and Enterprise Systems Engineering
Department
Massimo Fornasier
Technical University of Munich
Department of Mathematics
The annual LIDS Student Conference is a student-organized, studentrun event that provides an opportunity for graduate students to present their research to peers and the community at large. The conference also features a set of distinguished plenary speakers. The 2021 Student Conference marks 24 years of this signature lab event.
Student Speakers
Anish Agarwal
Anurag Ajay
Jason Altschuler
Juncal Arbelaiz
Sitan Chen
Yingnan Cui
Shuvomoy Das Gupta
Josh Gruenstein
Anubhav Guha
Chin-Chia Hsu
Eren C. Kizildag
Max Li
Parker Lusk
Tobia Marcucci
Hanzhang Qin
Bharadwaj Satchidan-
andan
Muhammed Sayin
Micah Smith
Chandler Squires
Behrooz Tahmasebi
Yi Tian
Lei Xu
Yunzong Xu
Chenyang Yuan
Panelists
Prof. Abhay Parekh
University of California, Berkeley
Prof. Devavrat Shah
MIT
Prof. Caroline Uhler
MIT
Dr. Kalyan Veeramachaneni
MIT
Plenary Speakers
Kimon Drakopoulos
(University of Southern California)
Zico Kolter
(Carnegie Mellon University)
Student Conference
Chairs
Manon Revel
Manxi Wu
Heng Yang
Mark York
NEW LIDS FACULTY
David Simchi-Levi
David Simchi-Levi is a professor in the MIT Department of Civil and Environmental Engineering, a core faculty member of IDSS, and serves as the head of the MIT Data Science Lab. He is considered one of the premier thought leaders in supply chain management and business analytics.
David is the current Editor-in-Chief of Management Science, one of the two flagship journals of INFORMS. He served as the Editor-in-Chief for Operations Research (2006-2012), the other flagship journal of INFORMS and for Naval Research Logistics (2003-2005). He has received numerous prestigious awards for his work, including the 2020 INFORMS Impact Prize for playing a leading role in developing and disseminating a new highly impactful paradigm for the identification and mitigation of risks in global supply chains.
IN REMEMBRANCE OF MATTHEW BRENNAN (1994-2021)
It is with deep sadness that we share the loss of LIDS student Matthew Brennan, who passed away suddenly on January 26, 2021, as the result of a previously unknown medical condition.
A longtime member of the MIT community, Matt earned his B.Sc in 2016, and his SM degree in 2018, both from the Department of Electrical Engineering and Computer Science (EECS). For his Master’s thesis, he was awarded the Ernst A. Guillemin Award for Best Thesis in Electrical Engineering. Matt was awarded his Ph.D., also from EECS, posthumously in May 2021.
Although still in the very beginning of his career, Matt was a true rising star. In 2018 his paper, “Reducibility and Computational Lower Bounds for Problems with Planted Sparse Structure” was given the Best Student Paper Award at the Conference on Learning Theory (COLT), the leading conference on the theory of machine learning. In 2020, he was again awarded the Best Student Paper Award at COLT for a related work. In the Fall of 2021 Matt was to have started a prestigious Miller Fellowship for postdoctoral work at UC Berkeley. Established members of the international research community at the interface of statistics and computation described Matt’s doctoral work as ground-breaking, opening the door to exciting lines of inquiry that will continue to be explored for years to come. Matt’s love of learning was boundless, and he was fiercely dedicated to hard work.
But mostly Matt was passionate about his wonderful friends, the tight-knit group from Upper Canada College, his many friends from the Olympiad math contest world, and his many friends from MIT. He was an extremely generous, kind, optimistic and enthusiastic person with an infectious laugh and a love of storytelling.
Matt is sorely missed by all who knew him—in the LIDS community and beyond. As we mourn this loss, we strive, in the words of MIT President L. Rafael Reif in his message to the MIT community, to “take inspiration from Matthew’s marvelous spirit, brilliant mind and generous heart.”