Friday, April 12 – E15-070 – Bartos Theatre

9:00am – Registration and Breakfast

9:25am – Opening Remarks
Ankur Moitra, Director, SDSC

9:30am – Plenary Speaker
Philippe Rigollet, Professor, MIT Mathematics
“Transformers are Clustering Machines”

Since their introduction in 2017, Transformers have revolutionized large language models and the broader field of deep learning. Central to this success is the groundbreaking self-attention mechanism. In this presentation, I’ll introduce a mathematical framework that casts this mechanism as a mean-field interacting particle system, revealing a desirable long-time clustering behavior. Joint work with Borjan Geshkovski, Cyril Letrouit, and Yury Polyanskiy.

10:15am – Break

10:30am – Student and Postdoc Lightning Talks

11:15am – Plenary Speaker
Sendhil Mullainathan, Roman Family University Professor of Computation and Behavioral Science, University of Chicago
“Machine Learning for Scientific Discovery”

Machine learning algorithms can find predictive signals that researchers fail to notice; yet they are notoriously hard-to-interpret.  I describe techniques for extracting novel theoretical insights from such black-boxes; and illustrate with applications to better understanding human behavior.

12:00pm – Lunch

1:00pm – Plenary Speaker
Vladimir Koltchinskii, Georgia Institute of Technology
“Functional estimation, bootstrap chains and high-dimensional Gaussian approximation”

In high-dimensional statistical inference, it is important to estimate low dimensional features of high-dimensional models. Such features are often represented by non-linear functionals of high-dimensional parameters and the design of their optimal statistical estimators is a challenging open problem. We will discuss an approach to this problem based on iterative higher order bias reduction and show how to obtain sharp bounds on the accuracy of resulting estimators of smooth functionals in the cases when there exists an estimator of the parameter itself for which Gaussian approximation holds in high dimensions. We will also briefly discuss the connection of this problem to high-dimensional central limit theorems. 

1:45pm – Closing Remarks
Ankur Moitra, Director, SDSC