Skip to content. Skip to navigation
CIM Menus

Informal Systems Seminar (ISS)

Distributed Stochastic Convex Optimization

Michael Rabbat, Associate Professor
Department of Electrical and Computer Engineering McGill University

November 7, 2014 at  10:30 AM
McConnell Engineering Room 437

This talk considers the problem of distributed convex optimization in a stochastic setting. Each node in a network of processors has a stochastic oracle for a common objective function, and the aim of the network is to collectively minimize the objective as quickly as possible. Such a problem arises, e.g., in large-scale machine learning where the goal of the network is to fit a model to training data that is spread across multiple nodes. We study a consensus-based approach where nodes individually take descent steps and then consensus iterations are performed to synchronize models across the nodes. We prove that the proposed method achieves the optimal centralized regret bound when the objective function has Lipschitz continuous gradients, and we discuss the tradeoff between communication, computation, and the network topology. This is joint work with Konstantinos Tsianos.