Best of Both Worlds in Federated LSA: Speedup When Possible, Personalization Always

2026-10-08

AI Summary

This study addresses the challenge in personalized federated linear stochastic approximation where a single algorithm struggles to simultaneously achieve acceleration and personalization. To this end, we propose PF-LSA, which introduces a novel adaptive mechanism requiring no prior knowledge of heterogeneity, optimally balancing local and globally averaged updates through mixing. Theoretically, leveraging a sharp decomposition of consensus and divergence errors, we prove that PF-LSA attains linear speedup under low heterogeneity while preserving personalization under high heterogeneity. Empirically, PF-LSA consistently converges to personalized solutions across all scenarios and significantly accelerates convergence on similar tasks without incurring additional computational overhead.

Abstract

We study personalized federated linear stochastic approximation (LSA), a framework which notably encompass personalized temporal difference learning. In this setting, heterogeneous agents collaborate to solve distinct linear fixed-point equations, each corresponding to an agent-specific learning problem. A central open question in personalized learning is whether a single method can adapt to an unknown level of heterogeneity by converging to each agent's personalized solution in all regimes while achieving a linear speedup in the number of agents when their learning problems are sufficiently similar. We answer this question affirmatively by introducing PF-LSA, a minimalist algorithm that mixes each agent's local stochastic update with the average update across agents, at no additional computational cost relative to standard federated methods. We prove that PF-LSA, achieves best-of-both-worlds guarantees without any prior knowledge on the level of heterogeneity. Our analysis is based on a sharp decomposition of the error into consensus and disagreement components. The consensus error decays rapidly, whereas the disagreement error decays more slowly but becomes negligible in low-heterogeneity regimes.

Problem

• Federated Learning
• Personalization
• Linear Stochastic Approximation
• Heterogeneity
• Linear Speedup

Innovation

• Personalized Federated Learning
• Linear Stochastic Approximation
• Heterogeneity Adaptation
• Error Decomposition
• Linear Speedup