Comments on “VCD-FL: Verifiable, Collusion-Resistant, and Dynamic Federated Learning”

Abstract

Gao et al. (2023)recently proposed a collusion-resistant and verifiable federated learning framework named VCD-FL (IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, vol. 18, pp. 3760–3773, 2023). However, in this letter, we show that VCD-FL fails to achieve its claimed security goals. In particular, we demonstrate that their designed commitment scheme, which serves as the core component of the proposed collusion-resistant verification mechanism, is unsafe, and then we present a feasible collusion attack launched by the aggregation server and corrupt clients by leveraging the existing security vulnerabilities.