Abnormal Client Behavior Detection in Federated Learning

TL;DR

This paper introduces a server-side method for detecting abnormal clients in federated learning by generating low-dimensional surrogates of model weights, effectively identifying malicious or malfunctioning clients to enhance system robustness.

Contribution

The paper proposes a novel server-side anomaly detection approach using low-dimensional surrogates of model weights in federated learning, outperforming traditional defense methods.

Findings

Significantly better detection accuracy than conventional methods

Effective identification of malicious and malfunctioning clients

Validated on FEMNIST image classification dataset

Abstract

In federated learning systems, clients are autonomous in that their behaviors are not fully governed by the server. Consequently, a client may intentionally or unintentionally deviate from the prescribed course of federated model training, resulting in abnormal behaviors, such as turning into a malicious attacker or a malfunctioning client. Timely detecting those anomalous clients is therefore critical to minimize their adverse impacts. In this work, we propose to detect anomalous clients at the server side. In particular, we generate low-dimensional surrogates of model weight vectors and use them to perform anomaly detection. We evaluate our solution through experiments on image classification model training over the FEMNIST dataset. Experimental results show that the proposed detection-based approach significantly outperforms the conventional defense-based methods.