PASS: Parameters Audit-based Secure and Fair Federated Learning Scheme against Free Rider

Federated Learning (FL) as a secure distributed learning frame gains interest in Internet of Things (IoT) due to its capability of protecting private data of participants. However, traditional FL systems are vulnerable to attacks such as Free-Rider (FR) attack, which causes not only unfairness but also privacy leakage and inferior performance to FL systems. The existing defense mechanisms against FR attacks only concern the scenarios where the adversaries declare less than 50% of the total amount of clients. Moreover, they lose effectiveness in resisting selfish FR (SFR) attacks. In this paper, we propose a Parameter Audit-based Secure and fair federated learning Scheme (PASS) against FR attacks. The PASS has the following key features: (a) works well in the scenario where adversaries are more than 50% of the total amount of clients; (b) is effective in countering anonymous FR attacks and SFR attacks; (c) prevents from privacy leakage without accuracy loss. Extensive experimental results verify the data protecting capability in mean square error against privacy leakage and reveal the effectiveness of PASS in terms of a higher defense success rate and lower false positive rate against anonymous SFR attacks. Note in addition, PASS produces no effect on FL accuracy when there is no FR adversary.