Provable Reduction in Communication Rounds for Non-Smooth Convex Federated Learning

2025/03/27 by Palenzuela, Karlo, Dadras, Ali, Yurtsever, Alp +1
#FOS: Computer and information sciences #FOS: Mathematics #Machine Learning (cs.LG) #Optimization and Control (math.OC)

paper · doi:10.48550/arxiv.2503.21627

Abstract

Multiple local steps are key to communication-efficient federated learning. However, theoretical guarantees for such algorithms, without data heterogeneity-bounding assumptions, have been lacking in general non-smooth convex problems. Leveraging projection-efficient optimization methods, we propose FedMLS, a federated learning algorithm with provable improvements from multiple local steps. FedMLS attains an ε-suboptimal solution in O(1/ε) communication rounds, requiring a total of O(1/ε²) stochastic subgradient oracle calls.

Related

• FedPAGE: A Fast Local Stochastic Gradient Method for Communication-Efficient Federated Learning
• Improved Generalization Bounds for Communication Efficient Federated Learning
• FedChain: Chained Algorithms for Near-Optimal Communication Cost in Federated Learning
• Faster Non-Convex Federated Learning via Global and Local Momentum
• Federated Learning with Compression: Unified Analysis and Sharp Guarantees