RIPPLE in Still Water: Zero-Shot Clustering in Federated Learning with Wavelet Scattering Transform

2026-10-02

AI Summary

This study addresses client drift caused by non-IID data in federated learning, alongside the high communication overhead and privacy risks inherent in existing clustering methods, by proposing the RIPPLE framework. RIPPLE introduces a novel offline spectral feature clustering mechanism based on the wavelet scattering transform, integrating variance-weighted principal component prototypes with a Gaussian mixture variational autoencoder. This design enables zero-shot clustering without gradient exchange during training and supports single-forward-pass personalized model generation for unseen clients. Theoretically, RIPPLE incurs communication costs comparable to FedAvg while achieving optimal convergence guarantees. Empirical evaluations across five benchmark datasets demonstrate that RIPPLE significantly outperforms existing baselines.

Abstract

Clustered Federated Learning (FL) partitions a client population into groups of similar local distributions and trains one specialized model per cluster, mitigating client drift that degrades single-model methods under non-IID data. Prior methods discover cluster structure inside the training loop through gradient similarity, loss evaluation, or EM-style updates, thus increasing communication overhead, exposing gradients to inversion attacks, and providing no mechanism to assign clients absent from training. We propose RIPPLE, a clustered FL framework in which cluster assignment is computed entirely offline from a spectral characterization of each client's local data: a variance-weighted principal-component prototype embedded via the Wavelet Scattering Transform and decoded by a Gaussian Mixture VAE trained server-side on synthetic client populations before federation begins. Per-round communication cost matches FedAvg exactly, and a client absent from training obtains a personalized model from a single forward pass, without gradient computation, model evaluation, or extra communication round. We prove that the gap between RIPPLE's surrogate clustered objective and the oracle is bounded by a computable quantity decaying with client sample size and independent of federation duration; per-cluster convergence matches the minimax-optimal rate for non-convex smooth objectives. Across five benchmarks spanning controlled and realistic heterogeneity, RIPPLE consistently outperforms all baselines, with margins growing on the most realistic partitions.

Problem

• Clustered Federated Learning
• Non-IID Data
• Client Drift
• Communication Overhead
• Privacy

Innovation

• Clustered Federated Learning
• Zero-Shot Clustering
• Wavelet Scattering Transform
• Gaussian Mixture VAE
• Offline Client Assignment