Detection and Prevention Against Poisoning Attacks in Federated Learning

TL;DR

This paper introduces a novel method called AADD for detecting and preventing poisoning attacks in federated learning by monitoring accuracy deviations and blacklisting malicious clients, thereby safeguarding the global model.

Contribution

It presents a new accuracy deviation detection approach (AADD) that effectively identifies and blocks poisoned clients in federated learning systems.

Findings

AADD accurately detects poisoned clients

Prevents deterioration of global model accuracy

Effective in various poisoning attack scenarios

Abstract

This paper proposes and investigates a new approach for detecting and preventing several different types of poisoning attacks from affecting a centralized Federated Learning model via average accuracy deviation detection (AADD). By comparing each client's accuracy to all clients' average accuracy, AADD detect clients with an accuracy deviation. The implementation is further able to blacklist clients that are considered poisoned, securing the global model from being affected by the poisoned nodes. The proposed implementation shows promising results in detecting poisoned clients and preventing the global model's accuracy from deteriorating.