On Distributed Online Classification in the Midst of Concept Drifts

TL;DR

This paper investigates the performance of distributed online learning algorithms in environments with changing data distributions, providing theoretical bounds and demonstrating the benefits of diffusion strategies over isolated learners.

Contribution

It offers new theoretical bounds for excess-risk in distributed online learning under concept drift and compares diffusion strategies to non-cooperative methods.

Findings

Diffusion strategies outperform non-cooperative learners in non-stationary environments.

Derived bounds quantify the generalization ability of distributed algorithms.

Simulations validate the theoretical results.

Abstract

In this work, we analyze the generalization ability of distributed online learning algorithms under stationary and non-stationary environments. We derive bounds for the excess-risk attained by each node in a connected network of learners and study the performance advantage that diffusion strategies have over individual non-cooperative processing. We conduct extensive simulations to illustrate the results.