Back to Home
Research Highlight 22 July 2026

Book on Stochastic Algorithms for Nonconvex Optimization and Reinforcement Learning by Prof. Vidyasagar

At a first glance, nonconvex optimization and Reinforcement Learning would appear to be quite disparate subjects. The unifying theme, as I bring out in the book, is that a method called "Stochastic Approximation" can be used to address both areas. In particular, I study Stochastic Gradient Descent (SGD), which is by now the default method used to train large neural networks, establish rates of convergence, as well as "optimal" learning rates.

Prof. M. Vidyasagar