Towards a Distributed Federated Learning Aggregation Placement using Particle Swarm Intelligence

TL;DR

The paper addresses efficient aggregation placement in Hierarchical Semi-Decentralized Federated Learning (SDFL) by employing a black-box Particle Swarm Optimization (PSO) approach that minimizes Total Processing Delay (TPD) without sharing internal client metrics. It introduces a PSO-based method where each particle encodes an aggregation configuration, and a fitness function based on $TPD$ guides optimization via standard PSO updates, including velocity and position equations. Through simulations across varying depths and widths and a docker-based real-world deployment on SDFLMQ, the approach demonstrates faster convergence and substantial reductions in total processing time compared to random or uniform baselines, validating the practicality of PSO for scalable edge federated learning. The work highlights the potential of light-weight, privacy-preserving optimization in distributed FL topologies and points to future enhancements in adapting PSO to dynamic topologies and richer fitness criteria.

Abstract

Federated learning has become a promising distributed learning concept with extra insurance on data privacy. Extensive studies on various models of Federated learning have been done since the coinage of its term. One of the important derivatives of federated learning is hierarchical semi-decentralized federated learning, which distributes the load of the aggregation task over multiple nodes and parallelizes the aggregation workload at the breadth of each level of the hierarchy. Various methods have also been proposed to perform inter-cluster and intra-cluster aggregation optimally. Most of the solutions, nonetheless, require monitoring the nodes' performance and resource consumption at each round, which necessitates frequently exchanging systematic data. To optimally perform distributed aggregation in SDFL with minimal reliance on systematic data, we propose Flag-Swap, a Particle Swarm Optimization (PSO) method that optimizes the aggregation placement according only to the processing delay. Our simulation results show that PSO-based placement can find the optimal placement relatively fast, even in scenarios with many clients as candidates for aggregation. Our real-world docker-based implementation of Flag-Swap over the recently emerged FL framework shows superior performance compared to black-box-based deterministic placement strategies, with about 43% minutes faster than random placement, and 32% minutes faster than uniform placement, in terms of total processing time.

Figures (4)

• Figure 1: Overview of Parameter sharing for aggregation using Pub/Sub communication in a Clustered Semi-Decentralized Federated Learning Topology.
• Figure 2: Proposed PSO-based aggregation placement in SDFL.
• Figure 3: Simulation results of PSO optimization in aggregation placement in SDHFL.
• Figure 4: Comparing aggregation placement using Random, PSO-based, and Round-Robin-based placement in SDFLMQ