\mathsf{VISTA}: Decentralized Machine Learning in Adversary Dominated Environments

2026-05-08

AI Summary

This work addresses the challenge of decentralized machine learning under adversarial settings where adversaries control a majority of worker nodes, rendering conventional robust aggregation methods ineffective due to their reliance on an honest majority assumption. The paper proposes VISTA, a novel framework that achieves asymptotic convergence without requiring an honest majority. VISTA models adversaries as rational agents and introduces a history-aware dynamic consensus threshold mechanism, compelling them to balance between injecting errors and facing rejection risks. By integrating incentive-compatible design with adaptive threshold tuning, VISTA optimizes the decentralized stochastic gradient descent (SGD) process. Theoretical analysis establishes that VISTA preserves the same asymptotic convergence guarantees as standard SGD, while empirical evaluations demonstrate its significant superiority over fixed-threshold approaches.

Abstract

Decentralized machine learning often relies on outsourcing computations, such as gradient evaluations, to untrusted worker nodes. Existing robust aggregation methods can mitigate malicious behavior under honest-majority assumptions, but may fail when adversaries control a majority of the workers. We study this adversary-dominated setting through an incentive-oriented framework in which reports are accepted and rewarded only when they are mutually consistent up to a threshold. This turns the adversary from a pure saboteur into a rational agent that trades off increasing estimation error against the risk of rejection and loss of reward. We consider iterative optimization under this model. Unlike one-shot computation, iterative learning requires long-horizon decisions: permissive acceptance rules enable faster early progress but admit more adversarial corruption, while strict rules improve estimation accuracy but cause frequent rejections. We propose \mathsf{VISTA}, an adaptive algorithm that tunes the acceptance threshold using the optimization history. Numerical results show that \mathsf{VISTA} improves convergence over static thresholds. We also provide a rigorous convergence analysis showing that, with suitable incentive-aware adaptation, adversary-dominated decentralized learning can retain the asymptotic convergence behavior of standard SGD without relying on an honest majority.

Problem

• decentralized machine learning
• adversary-dominated environments
• robust aggregation
• honest-majority assumption
• convergence

Innovation

• decentralized learning
• adversary-dominated environments
• incentive mechanism
• adaptive thresholding
• robust aggregation