News

Guy Bresler, Tamara Broderick, Stefanie Jegelka, Suvrit Sra

Four SDSC faculty awarded promotions

February 07, 2019

The Statistics and Data Science Center is delighted to share that four faculty members have been promoted to the rank of associate professor without tenure, effective July 1, 2019: Guy Bresler, Tamara Broderick, Stefanie Jegelka, and Suvrit Sra. All are EECS faculty and core SDSC members.

Guy Bresler received his PhD in EECS from the University of California at Berkeley in 2012. After a post-doctoral fellowship at MIT, he became assistant professor in EECS and a core faculty member of IDSS in 2015. Guy’s research focuses on high-dimensional statistical inference. His work considers two key factors for an inference or learning task to be possible: (1) informational/statistical complexity (does the data contain enough information for the task to be in principle feasible?), and (2) computational complexity (does the problem have a structure that can be exploited in order to obtain computationally feasible algorithms?). He has made several central contributions to this field. Recent work addressed the question of “complexity” of statistical problems. This work is, in many ways, an “average complexity” analog of the traditional field of worst-case complexity in theoretical computer science. In groundbreaking work, Guy developed a comprehensive theory that maps out a rich web of relations between important statistical problems such as sparse PCA, community detection, bi-clustering, and others, ultimately showing that these problems are at least as hard as the problem of discovering a planted clique in a random graph. In statistics, the planted clique problem is accepted as a computationally intractable problem (within a certain parameter regime), which is analogous to what happens in traditional complexity theory where NP-hard problems are accepted (conjectured) to be intractable.

Guy has taught and contributed to the curricula of multiple EECS classes, including probability theory, machine learning, algorithms for inference, and discrete stochastic processes. He mentors several PhD students; a student of his won the competitive best paper award in the Conference on Learning Theory (COLT). He serves as co-chair of the ML/INFR student admission group in EECS, and organizes external seminars in both LIDS and SDSC.

Tamara Broderick received her PhD from UC Berkeley in 2014 and joined the MIT EECS faculty in January 2015. Tamara works in machine learning and statistics. Her research focuses on Bayesian inference and graphical models — with an emphasis on scalable and nonparametric methods. She has made many significant and foundational contributions to her fields. For instance, her work demonstrates how to summarize data as a pre-processing step before running Bayesian, and other, machine learning methods. Her methods are fast and practical, come with theoretical guarantees, and are already widely used. Her work also demonstrates how to quickly and automatically calculate a local form of sensitivity to assess modeling assumptions in Bayesian analysis. This work simultaneously provides dramatically improved uncertainty quantification for popular Bayesian approximations.

Tamara has received an NSF CAREER Award, a Sloan Research Fellowship, an Army Research Office Young Investigator Award, a Google Faculty Research Award, an International Society for Bayesian Analysis Lifetime Members Junior Research Award, and a Marshall Scholarship. In addition, she received the Jerome H. Saltzer Award for Excellence in Teaching in recognition of her phenomenal teaching ability. She is also a sought-after lecturer, having given more than 100 invited talks, seminars, and tutorials worldwide. Tamara is an affiliate faculty member of IDSS.

Stefanie Jegelka received her PhD in computer science in 2012 from ETH Zurich and the Max Planck Institute for Intelligent Systems. After serving as a postdoctoral researcher at UC Berkeley, she joined MIT EECS as an assistant professor in early 2015. Stefanie’s work span the theory and practice of algorithmic machine learning and optimization. She is considered a leader in submodular optimization, a field that has been extremely important for computer vision and machine learning to accommodate problems with discrete combinatorial structures. Her work combines deep theoretical understanding with practical motivation and efficient implementation, providing algorithms with exceptional practical performance and rigorous theoretical guarantees for several key problems. Her current work also studies negative association in discrete probability and machine learning (which enables diverse selection of data points) and discrete probabilistic inference and Bayesian optimization. In addition, Stefanie has made significant teaching contributions to the department. She developed a graduate course, “Learning with Combinatorial Structure,” that covers models, algorithms, and applications, analyzing how various types of mathematical structures can be used for machine learning. She also co-developed (with Caroline Uhler) a new hands-on data analysis course, “Statistics, Computation, and Applications.” She teaches several other courses, including 6.862 (Applied Machine Learning), 6.437 (Inference and Information), and 6.046 (Design and Analysis of Algorithms).

An active member of the machine learning community, Stefanie has organized several workshops on optimization and combinatorial machine learning, given invited tutorials, and has been on the senior program committee for several conferences, including NIPS, ICML, AISTATS, and UAI. She has received an NSF CAREER Award, a DARPA Young Faculty Award, the German Pattern Recognition Award, and a Sloan Research Fellowship. Stefanie is an affiliate faculty of IDSS.

Suvrit Sra received his PhD in computer science from the University of Texas at Austin in 2007. He was a senior research scientist at the Max Planck Institute for Intelligent Systems and a principal research scientist in LIDS before he joined the EECS faculty in 2018. He is also a core faculty member of IDSS.

Suvrit is a leading researcher in the active area of optimization for machine learning. His research bridges areas such as optimization, matrix theory, geometry, and probability with machine learning. Convex optimization methods, in particular stochastic gradient descent (SGD) and its variants, have become the standard workhorse in this area with much recent progress. Suvrit has made fundamental contributions to both convex and nonconvex optimization for machine learning, including developing variance reduction techniques for nonconvex problems with significant improvement over SGD and finding tractable classes of nonconvex optimization problems by exploring the underlying geometry of the problem and discovering the “hidden convexity” in important problems. More broadly, he is interested in data-driven questions within engineering, science, and health care.

Suvrit has contributed significantly to the department’s educational mission by teaching both graduate and undergraduate courses in machine learning. He is also active in the larger machine learning community. He founded the OPT (Optimization for Machine Learning) series of workshops at the NIPS conference, which he has co-chaired since 2008; he has also edited a popular book with the same title, published by MIT Press in 2011. His work has won several honors at machine learning venues, including the 2011 SIAM Outstanding Paper Prize.