DEFEND: A Poisoned Model Detection and Malicious Client Exclusion Mechanism for Secure Federated Learning Road Condition Classification
DEFEND: Poisoned Model Detection and Malicious Client Exclusion Mechanism for Secure Federated Learning-based Road Condition Classification
AI summary
DEFEND detects poisoned models through per-neuron magnitude analysis and Gaussian mixture model (GMM) clustering, and in each round discards their contributions and adjusts client scores, thereby excluding malicious clients. Across multiple federated learning road condition classification models and tasks, it defends against targeted label flipping attacks, outperforming seven baselines by at least 15.78%. Performance under attack is the same as in the no-attack scenario.