An Enhanced Combinatorial Contextual Neural Bandit Approach for Client Selection in Federated Learning

Xiangyu Ma, Wei Shi, Junfeng Wen

Abstract: In the evolving landscape of machine learning (ML), federated learning (FL) stands out as an innovative strategy for training models across dispersed devices without centralizing raw data. Such an approach, however, grapples with data heterogeneity challenges, violating the independent and identically distributed (IID) assumption and undermining the global model accuracy. To address this, we present federated adversary-resilient neural selector (FANS), a sophisticated context-aware client selection algorithm, leveraging a combinatorial contextual neural bandit framework. This algorithm that accentuates the enhanced extraction of contextual information by evaluating each local client with a universally standardized dataset, subsequently yielding a more insightful contextual representation tailored for federated settings. In addition, we introduce selection robustness score (SRS), a novel metric designed to quantify the efficacy of client selection in the presence of adversarial conditions. Using this metric, we demonstrate FANS’s effectiveness in enhancing the FL process. Empirical evaluations across diverse settings reveal our method’s superiority over current state-of-the-art solutions, with significant improvements in both SRS and global model accuracy.