FedCCA: Client-Centric Adaptation against Data Heterogeneity in Federated Learning on IoT Devices

TL;DR

FedCCA tackles data heterogeneity in Federated Learning for IoT by introducing a client-centric adaptation framework that personalizes models per client via a dedicated client-specific encoder and dynamic client selection. It employs an attention-based global aggregation to transfer knowledge from multiple similar clients and updates local models through a weighted combination of peer models. Empirical results across multiple digit and object-recognition datasets show FedCCA outperforming strong baselines, with ablations confirming the importance of multi-source aggregation and client-centric selection. The work demonstrates a practical, privacy-preserving approach to personalized FL in heterogeneous IoT deployments and points to extensions for broader edge computing scenarios.

Abstract

With the rapid development of the Internet of Things (IoT), AI model training on private data such as human sensing data is highly desired. Federated learning (FL) has emerged as a privacy-preserving distributed training framework for this purpuse. However, the data heterogeneity issue among IoT devices can significantly degrade the model performance and convergence speed in FL. Existing approaches limit in fixed client selection and aggregation on cloud server, making the privacy-preserving extraction of client-specific information during local training challenging. To this end, we propose Client-Centric Adaptation federated learning (FedCCA), an algorithm that optimally utilizes client-specific knowledge to learn a unique model for each client through selective adaptation, aiming to alleviate the influence of data heterogeneity. Specifically, FedCCA employs dynamic client selection and adaptive aggregation based on the additional client-specific encoder. To enhance multi-source knowledge transfer, we adopt an attention-based global aggregation strategy. We conducted extensive experiments on diverse datasets to assess the efficacy of FedCCA. The experimental results demonstrate that our approach exhibits a substantial performance advantage over competing baselines in addressing this specific problem.

Figures (7)

• Figure 1: Illustrative examples of heterogeneous scenarios in federated learning on IoT devices.
• Figure 2: FedCCA algorithm overview. At client-side, each client mainly consists of two components: a local model $\theta$, a client-specific model $\phi$ to capture client information. At server-side, we adopt a client-centric selection strategy to sample the optimal client set for each client based on the client-specific representations extracted from client-specific model $\phi$ in the previous step.
• Figure 3: Illustration of client-centric adaptation in FedCCA.
• Figure 4: Training performance over CIFAR-100 under different epochs.
• Figure 5: Client selection counts over different datasets.