Study Examines Temporal Attacks in Federated Learning Systems
/ 1 min read
🕒🔍 Temporal attacks pose significant threats to Federated Learning systems. A recent study investigates the impact of adversarial clients on the performance of various Federated Learning (FL) models, including Multinominal Logistic Regression, Random Forest, and several Neural Network architectures. The findings reveal that temporal attacks, particularly when adversaries are active during critical rounds, severely degrade model performance. The research underscores the necessity for enhanced strategies to bolster the robustness of FL processes against such attacks. Additionally, the paper briefly evaluates potential defense mechanisms, such as outlier detection within the aggregation algorithm, to mitigate these vulnerabilities.
