LEAP: Optimization Hierarchical Federated Learning on Non-IID Data with Coalition Formation Game

TL;DR

This work tackles non-IID data and communication bottlenecks in Hierarchical Federated Learning by proposing LEAP, which combines a coalition-formation game to align edge data distributions with a gradient-projection-based bandwidth optimization. The method decomposes the problem into data-distribution alignment, bandwidth allocation, and power control, proving the existence of stable coalitions and delivering closed-form power strategies under latency constraints. Empirically, LEAP yields significant improvements in model accuracy (up to $20.62\%$) and substantial energy savings ($\ge 2.24\times$) across multiple real-world datasets, highlighting its practical potential for IoT edge networks. Overall, LEAP provides a scalable, privacy-preserving mechanism to optimize multi-dimensional performance in HFL without sharing raw data."

Abstract

Although Hierarchical Federated Learning (HFL) utilizes edge servers (ESs) to alleviate communication burdens, its model performance will be degraded by non-IID data and limited communication resources. Current works often assume that data is uniformly distributed, which however contradicts the heterogeneity of IoT. Solutions of additional model training to check the data distribution inevitably increases computational costs and the risk of privacy leakage. The challenges in solving these issues are how to reduce the impact of non-IID data without involving raw data and how to rationalize the communication resource allocation for addressing straggler problem. To tackle these challenges, we propose a novel optimization method based on coaLition formation gamE and grAdient Projection, called LEAP. Specifically, we combine edge data distribution with coalition formation game innovatively to adjust the correlations between clients and ESs dynamically, which ensures optimal correlations. We further capture the client heterogeneity to achieve the rational bandwidth allocation from coalition perception and determine the optimal transmission power within specified delay constraints at client level. Experimental results on four real datasets show that LEAP is able to achieve 20.62% improvement in model accuracy compared to the state-of-the-art baselines. Moreover, LEAP effectively reduce transmission energy consumption by at least about 2.24 times.

Figures (7)

• Figure 1: An overview of LEAP
• Figure 2: Changes of data distribution and $\mathcal{\overline{JS}}$ during coalition formation.
• Figure 3: Global model performance comparison of different data distributions and methods on four datasets.
• Figure 4: Transmission energy consumption and transmission power under different optimization schemes.
• Figure 5: Variations in data distribution across three different methods.

Theorems & Definitions (8)

• Definition 1
• Definition 2
• Definition 3
• Definition 4
• Definition 5
• Theorem 1
• Lemma 1
• Theorem 2