Tommi S. Jaakkola, Ph.D. Thomas Siebel Professor of Electrical Engineering and Computer Science and the Institute for Data, Systems, and Society MIT Computer Science and Artificial Intelligence Laboratory Stata Center, Bldg 32-G470 Cambridge, MA 02139 tommi at csail dot mit dot edu [home]   [papers]   [research]   [courses]   [people]

Short summaries of some on-going projects

Generative design of biomolecules Generative AI offers considerable potential to advance molecular therapeutics. Our work focuses on developing foundational generative tools for the inverse design of proteins, nanobodies, and other therapeutic modalities. Some of the key technical challenges that we address include steering models toward binding selectivity and other prescribed characteristics as well as incorporating physics more integrally into the models. Diffusion-based generative models learn to emulate physics largely from examples of existing or predicted complexes. As such these models do not yet capture the full conformational flexibility, diversity or binding characteristics of real proteins. We explore ways of incorporating physics/physical simulations into the model training / generation as well as developing generative models to accelerate simulations themselves.

Molecular optimization Drug design relies increasingly on the ability to automatically optimize molecules towards better biochemical or biomedical properties. Our goal in this context is to accelerate and enable inverse molecular design by developing methods that can programmatically transform a precursor molecule into a refined version that satisfies user-specified property characteristics. Technical challenges in this context involve multi-resolution molecular representations and the ability to realize novel molecular structures as predictions.

Language models, reasoning We explore new approaches to multi-resolution, multi-modal language modeling, including coarse graining and effective integration with physical characteristics. For example, we develop models that reason directly over evidence rather than treating measurements as weakly coupled side information. More broadly, we study coordinated reasoning as a general strategy for grounding and reconciling scientific inferences.

Strategic modeling/control We explore classes of structured prediction models where the resolution of the predicted outcome involves significant strategic interactions. Our game theoretic models map the input context to utilities and interactions, and guides the game dynamics into a near equilibrium. We explore different types of game theoretic models, associated dynamics, and theoretical guarantees of convergence, identifiability, and generalization. In particular, we focus on coordination games designed to reconcile multiple potentially conflicting views/roles/models towards a coherent outcome.