Evaluating Gradient Inversion Attacks and Defenses in Federated Learning
This work addresses privacy threats in federated learning for practitioners, but it is incremental as it builds on existing attacks and defenses.
The paper evaluates gradient inversion attacks and defenses in federated learning, finding that relaxing attack assumptions weakens them and that combining defenses reduces effectiveness with minor data utility loss, such as a 5% drop in model accuracy under certain conditions.
Gradient inversion attack (or input recovery from gradient) is an emerging threat to the security and privacy preservation of Federated learning, whereby malicious eavesdroppers or participants in the protocol can recover (partially) the clients' private data. This paper evaluates existing attacks and defenses. We find that some attacks make strong assumptions about the setup. Relaxing such assumptions can substantially weaken these attacks. We then evaluate the benefits of three proposed defense mechanisms against gradient inversion attacks. We show the trade-offs of privacy leakage and data utility of these defense methods, and find that combining them in an appropriate manner makes the attack less effective, even under the original strong assumptions. We also estimate the computation cost of end-to-end recovery of a single image under each evaluated defense. Our findings suggest that the state-of-the-art attacks can currently be defended against with minor data utility loss, as summarized in a list of potential strategies. Our code is available at: https://github.com/Princeton-SysML/GradAttack.