proximal algorithms

The Broximal Point Method and Its Impact on Optimization Theory and Machine Learning

Peter Richtarik, Professor, Computer Science

Sep 21, 12:00 - 13:00

B9 R2325

Broximal Point Method proximal algorithms machine learning gradient methods

This talk introduces the Broximal Point Method (BPM), a condition-number-free alternative to classical proximal methods, that replaces proximal penalties with ball constraints, surveys its theoretical extensions and distributed variants, and explains how approximate non-Euclidean Brox updates underpin practical machine learning optimizers.

The Many Faces of Compression: Theory and Practice in Federated Optimization

Elnur Gasanov, Ph.D. Student, Computer Science

May 28, 10:00 - 12:00

B3, L5, R5209

compression methods communication efficiency proximal algorithms

This thesis presents novel compression methods and optimization algorithms aimed at improving communication efficiency in large-scale machine learning systems.