Adversary-Robust Learning from Fully Asynchronous Directional Derivative Estimates

TL;DR

FAR-SIGN is a novel asynchronous learning algorithm that uses sign-based updates for robustness against adversaries, with proven convergence and superior performance in experiments.

Contribution

It introduces FAR-SIGN, a fully asynchronous, adversary-resilient optimization method with convergence guarantees and improved efficiency over existing robust aggregation techniques.

Findings

FAR-SIGN converges almost surely to stationary points for nonconvex objectives.

It achieves near-optimal convergence rates in both first-order and zeroth-order settings.

Experiments on MNIST demonstrate superior accuracy and efficiency compared to existing methods.

Abstract

We propose FAR-SIGN (Fully Asynchronous Robust optimization via SIGNed directional projections) for adversary-resilient learning in parameter-server--worker systems. FAR-SIGN achieves robustness through sign-based updates along carefully designed directions and mitigates the resulting bias via a two-timescale mechanism. It admits both first-order and zeroth-order implementations and enables fully asynchronous execution without requiring a private reference dataset at the server. We establish almost-sure convergence of FAR-SIGN to the set of stationary points for smooth, nonconvex objectives. Moreover, we prove the near-optimal rate of $O(n^{-1/4+\epsilon})$ in the first-order setting and the standard $O(n^{-1/6+\epsilon})$ in the zeroth-order setting, where $n$ is the iteration count and $\epsilon>0$ can be chosen arbitrarily small. Experiments on MNIST show that FAR-SIGN outperforms robust aggregation-based methods in both accuracy and wall-clock time.