Systems | Information | Learning | Optimization
 

SILO: Universality in High-Dimensional Statistics

Abstract

Universality refers to the probabilistic phenomenon where the behavior of many large and complicated random systems can be accurately described by a simple and mathematically tractable model that is faithful to a few important properties of these systems.

The most well-known universality result is the central limit theorem, which states that sample averages computed on large datasets follow a universal distribution, namely the Gaussian distribution. This result forms the backbone of classical statistics, which studies low-dimensional statistical inference problems.

In this talk, I will discuss some of my past and ongoing work on understanding a new universality phenomenon observed in the context of high-dimensional statistical inference problems and how it can be exploited to obtain optimal estimation algorithms for these problems.

February 7, 2024
12:30 pm (1h)

Discovery Building, Researchers’ Link

Rishabh Dudeja