We work on a variety of topics spanning theoretical foundations, algorithms, and applications.
We study a range of research areas related to machine learning and their applications for robotics, health care, language processing, information retrieval and more. Among these subjects include precision medicine, motion planning, computer vision, Bayesian inference, graphical models, statistical inference and estimation. Our work is interdisciplinary and deeply rooted in systems and computer science theory.
Our goal is to develop methods that can "explain" the behavior of complex machine learning models, without restricting their power. We seek explanations that are simple, robust and grounded in statistical analysis of the model's behavior.
We aim to understand theory and applications of diversity-inducing probabilities (and, more generally, "negative dependence") in machine learning, and develop fast algorithms based on their mathematical properties.
Data often has geometric structure which can enable better inference; this project aims to scale up geometry-aware techniques for use in machine learning settings with lots of data, so that this structure may be utilized in practice.
Many optimization problems in machine learning rely on noisy, estimated parameters. Neglecting this uncertainty can lead to great fluctuations in performance. We are developing algorithms for these already nonconvex problems that are robust to such errors.
This project aims to uncover theoretical properties and new applications of perturbation models, a family of probability distributions for high dimensional structured prediction problems.
Many modern Bayesian models involve infinitely many latent parameters. We seek to develop finite approximations which are more tractable for use in practice, and characterize their incurred error.
Most popular, tractable statistical models for network data inherently assume the network is dense, although this is rarely true in practice; we propose a new modeling framework that correctly captures sparse networks.
To obtain scalable Bayesian inference methods, we develop algorithms to create compact “summaries” of large quantities of data. We can then quickly run standard inference algorithms on these summaries without needing to look at the whole dataset.
We develop statistical models that are prescriptive rather than predictive/descriptive. From an observational dataset, our methods learn to automatically identify beneficial actions that will improve outcomes, rather than requiring human-made decisions.
Our goal is to develop new tools for modeling diverse multi-agent settings, and design estimation algorithms to unravel the strategic interactions among the agents.
Our goal is to enable scalable and accurate Bayesian inference for rich probabilistic models by applying optimization techniques. We also aim to understand the connections between the two approaches of statistical inference: Bayesian and frequentist.
MIT EECS associate professors and CSAIL principal investigators Tamara Broderick and Michael Carbin received the Presidential Early Career Award for Scientists and Engineers (PECASE). Broderick was awarded for her work on robust machine learning, while Carbin was honored for his research on programming systems.
When you ask a robot to place blocks in matching bowls, for example, MIT’s “PRoC3S” method helps an LLM create a viable action plan by testing each step in a simulation. This strategy could eventually aid in-home robots to complete more ambiguous chore requests.
MIT CSAIL and Technical University of Munich researchers have proposed a new approach to self-supervised learning that addresses the limitations of relying on pre-defined data augmentations, and instead learning a general representation that can adapt to different transformations by paying attention to context.
New algorithm helps robots practice skills like sweeping and placing objects, potentially helping them improve at important tasks in houses, hospitals, and factories.
Stefanie Jegelka seeks to understand how machine-learning models behave, to help researchers build more robust models for applications in biology, computer vision, optimization, and more.
Every year 40,000 women die from breast cancer in the U.S. alone. When cancers are found early, they can often be cured. Mammograms are the best test available, but they’re still imperfect and often result in false positive results that can lead to unnecessary biopsies and surgeries.
