Abnormal Local Clustering in Federated Learning

TL;DR

This paper proposes a clustering method to identify abnormal local clients in federated learning by analyzing model similarity, enhancing the detection of anomalous data sources without compromising privacy.

Contribution

It introduces a novel Euclidean similarity clustering approach to distinguish normal and abnormal local clients in federated learning.

Findings

Effective separation of normal and abnormal clients achieved

Clustering method preserves privacy while detecting anomalies

Applicable to federated classification models

Abstract

Federated learning is a model for privacy without revealing private data by transfer models instead of personal and private data from local client devices. While, in the global model, it's crucial to recognize each local data is normal. This paper suggests one method to separate normal locals and abnormal locals by Euclidean similarity clustering of vectors extracted by inputting dummy data in local models. In a federated classification model, this method divided locals into normal and abnormal.