Temporal Analysis of Adversarial Attacks in Federated Learning
Rohit Mapakshi, Sayma Akther, Mark Stamp
TL;DR
This work analyzes the robustness of Federated Learning systems under adversarial clients with a focus on temporal attacks that unfold over training rounds. It systematically evaluates eight model families, including classical and neural networks, against data poisoning, model poisoning, and GAN reconstruction attacks using the Flower framework and FedAvg (with tree-based bagging for RF/XGBoost). Key findings show that attacks active in later rounds cause more damage, but simple defenses like outlier detection—especially One-Class SVM—can substantially mitigate harm, notably against model poisoning. The study highlights the need for stronger aggregation defenses and broader attack scenarios to ensure FL reliability in privacy-preserving, distributed settings.
Abstract
In this paper, we experimentally analyze the robustness of selected Federated Learning (FL) systems in the presence of adversarial clients. We find that temporal attacks significantly affect model performance in the FL models tested, especially when the adversaries are active throughout or during the later rounds. We consider a variety of classic learning models, including Multinominal Logistic Regression (MLR), Random Forest, XGBoost, Support Vector Classifier (SVC), as well as various Neural Network models including Multilayer Perceptron (MLP), Convolution Neural Network (CNN), Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM). Our results highlight the effectiveness of temporal attacks and the need to develop strategies to make the FL process more robust against such attacks. We also briefly consider the effectiveness of defense mechanisms, including outlier detection in the aggregation algorithm.