Adversary-Robust Learning from Fully Asynchronous Directional Derivative Estimates

2026-05-10

AI Summary

This work addresses the challenge of achieving robustness in distributed learning under fully asynchronous communication in a parameter server architecture while simultaneously defending against adversarial attacks. The authors propose FAR-SIGN, the first method that enables robust learning without requiring private reference data at the server. By integrating sign-based directional derivative estimation with a two-timescale update mechanism, FAR-SIGN supports both first-order and zeroth-order optimization and guarantees almost sure convergence for non-convex objectives. Theoretical analysis shows that its convergence rates are nearly optimal: $O(n^{-1/4+\varepsilon})$ for first-order and $O(n^{-1/6+\varepsilon})$ for zeroth-order settings. Empirical results on MNIST demonstrate that FAR-SIGN outperforms existing robust aggregation methods in both accuracy and wall-clock runtime.

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 $O(n^{-1/4+ε})$ in the first-order setting and the standard $O(n^{-1/6+ε})$ in the zeroth-order setting, where $n$ is the iteration count and $ε>0$ can be chosen arbitrarily small. Experiments on MNIST show that FAR-SIGN outperforms robust aggregation-based methods in both accuracy and wall-clock time.

Problem

• adversary-resilient learning
• fully asynchronous
• directional derivative estimates
• parameter-server systems
• nonconvex optimization

Innovation

• adversary-robust learning
• asynchronous optimization
• sign-based updates
• two-timescale mechanism
• zeroth-order optimization