Type-Based Unsourced Federated Learning With Client Self-Selection

TL;DR

This work tackles privacy-preserving client selection in over-the-air federated learning under data heterogeneity. It introduces a client self-selection mechanism where active clients decide participation by comparing their local loss to a centrally updated threshold, enabling privacy by keeping client identities hidden, and couples this with CSI-free Type-Based UMA (TUMA) for unsourced aggregation over a D-MIMO network. The approach integrates candidate-set formation, threshold-based participation, and AMP-based type estimation with vector-quantized updates, achieving test accuracy close to server-side PoC and outperforming random selection, while remaining robust to wireless impairments. The proposed CSI-free, unsourced framework holds practical potential for scalable, privacy-preserving FL in dense wireless deployments, with future work on parameter optimization and codebook design for further gains.

Abstract

We address the client-selection problem in federated learning over wireless networks under data heterogeneity. Existing client-selection methods often rely on server-side knowledge of client-specific information, thus compromising privacy. To overcome this issue, we propose a client self-selection strategy based solely on the comparison between locally computed training losses and a centrally updated selection threshold. Furthermore, to support robust aggregation of clients' updates over wireless channels, we integrate this client self-selection strategy into the recently proposed type-based unsourced multiple-access framework over distributed multiple-input multiple-output (D-MIMO) networks. The resulting scheme is completely unsourced: the server does not need to know the identity of the clients. Moreover, no channel state information is required, neither at the clients nor at the server side. Simulation results conducted over a D-MIMO wireless network show that the proposed self-selection strategy matches the performance of a comparable state-of-the-art server-side selection method and consistently outperforms random client selection.

Figures (4)

• Figure 1: Test accuracy as a function of the number $t$ of federated learning rounds for different selection strategies resulting in approximately the same average number of participating clients $K_\text{tar}=100$, under error-free communication. The actual average number of participating clients $|\mathcal{S}\xspace^{(t)}|$ for each selection strategy, as well as its standard deviation are reported in the legend.
• Figure 2: Test accuracy as a function of the number $t$ of federated learning rounds for the TUMA decoder with client self-selection; we consider vector quantization with parameters $J=7$, $Q=30$ and for different transmission blocklength values $N \in \{10, 20, 50\}$.
• Figure 3: Test accuracy as a function of the number $t$ of federated learning rounds; we compare TUMA and MD-AirComp with perfect CSI for $N=50$, $J=7$, $Q=30$, and client self-selection.
• Figure 4: Test accuracy achieved at round $t=100$ versus the number of quantization bits $J$ for TUMA decoder and client self-selection; $Q=30$.