Random Client Selection on Contrastive Federated Learning for Tabular Data
Achmad Ginanjar, Xue Li, Priyanka Singh, Wen Hua
TL;DR
Vertical Federated Learning enables privacy-preserving joint training on vertically partitioned data but remains susceptible to information leakage during intermediate computations. Contrastive Federated Learning mitigates some leakage by learning representations, yet gradient-based attacks threaten privacy and integrity; this work evaluates random client selection as a lightweight defense against such attacks in CFL. Across 10 real-world datasets, random client selection reduces attack success probabilities while preserving model performance, indicating robust defense with low computational overhead. Theoretical bounds such as $P(Attack) \le p_c * r_l$ and convergence bound $E[||\omega_t - \omega^*||^2] \le (1 - \eta * \mu * r_l)^t * ||\omega_0 - \omega^*||^2$ underpin the results and guide deployment of secure CFL in tabular data applications.
Abstract
Vertical Federated Learning (VFL) has revolutionised collaborative machine learning by enabling privacy-preserving model training across multiple parties. However, it remains vulnerable to information leakage during intermediate computation sharing. While Contrastive Federated Learning (CFL) was introduced to mitigate these privacy concerns through representation learning, it still faces challenges from gradient-based attacks. This paper presents a comprehensive experimental analysis of gradient-based attacks in CFL environments and evaluates random client selection as a defensive strategy. Through extensive experimentation, we demonstrate that random client selection proves particularly effective in defending against gradient attacks in the CFL network. Our findings provide valuable insights for implementing robust security measures in contrastive federated learning systems, contributing to the development of more secure collaborative learning frameworks