Detection and Isolation of Adversaries in Decentralized Optimization for Non-Strongly Convex Objectives

Abstract

Decentralized optimization has found a significant utility in recent years, as a promising technique to overcome the curse of dimensionality when dealing with large-scale inference and decision problems in big data. While these algorithms are resilient to node and link failures, they however, are not inherently Byzantine fault-tolerant towards insider data injection attacks. This paper proposes a decentralized robust subgradient push (RSGP) algorithm for detection and isolation of malicious nodes in the network for optimization non-strongly convex objectives. In the attack considered in this work, the malicious nodes follow the algorithmic protocols, but can alter their local functions arbitrarily. However, we show that in sufficiently structured problems, the method proposed is effective in revealing their presence. The algorithm isolates detected nodes from the regular nodes, thereby mitigating the ill-effects of malicious nodes. We also provide performance measures for the proposed method.