Multiplicative weights is a powerful and general algorithmic framework of wide use in various areas of computer science. In this talk, we will see their use in learning probability distributions, with a focus on efficiency in high dimensional graphical models. We will gently introduce the framework and show its connection to learning distributions, before applying them for learning distributions and compare the resulting algorithms with the state-of-the-art.
Bio: Sutanu Gayen is a faculty member in the CSE Department at IIT Kanpur. Sutanu's research lies broadly in algorithms and theory and specifically in theory of machine learning.
