Malicious gradient leakage attacks in federated learning are argued to be inherently limited and detectable in realistic settings, and a lightweight Z-score based detector is proposed.
Advances in Neural Information Pro- cessing Systems (NeurIPS)35, 7641–7654 (2022)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Hear No Evil: Detecting Gradient Leakage by Malicious Servers in Federated Learning
Malicious gradient leakage attacks in federated learning are argued to be inherently limited and detectable in realistic settings, and a lightweight Z-score based detector is proposed.