FedTopo: Relation-Level Topology Sharing for Model-Heterogeneous Federated Learning

2026-07-29

AI Summary

This work addresses the challenge of representation inconsistency in heterogeneous federated learning, where diverse local model architectures render conventional absolute-space knowledge sharing ineffective. To overcome this, the authors propose FedTopo, a novel framework that encodes global knowledge as category relation topologies, enabling alignment-free cross-client knowledge transfer by sharing relative structural relationships among categories rather than model parameters or prototypes. FedTopo integrates three key components: local prototype-based topology construction, a reliability-aware server-side aggregation mechanism, and a topology-similarity-guided negative class strategy for local training. Extensive experiments demonstrate that FedTopo significantly outperforms existing methods across three datasets and eight heterogeneous backbone architectures, achieving superior performance with low communication overhead and no additional inference cost.

Abstract

Federated learning (FL) enables collaborative learning over decentralized data silos without centralizing raw data. However, heterogeneous local architectures often induce non-aligned representation spaces, making it difficult to transfer global knowledge across silos. Existing paradigms share this knowledge as model parameters, distilled predictions, or class prototypes, yet all encode it in an absolute space that must be aligned across clients. Heterogeneous backbones break this alignment, so the shared knowledge becomes unreliable and misleads local training. We propose FedTopo, a relation-level framework that encodes global knowledge as class relation topology, capturing how classes relate within each client rather than where they lie in feature space. Each client builds its relation topology from local prototypes and uploads it with class statistics. The server then aggregates these relations in a reliability-aware manner that down-weights weakly supported ones, and broadcasts the global topology to clients. The global topology guides local training by emphasizing topology-similar negative classes. Experiments on three datasets under eight heterogeneous backbones show that FedTopo consistently outperforms parameter-, distillation-, and prototype-sharing baselines, with low communication and no inference overhead. Our code is available at https://github.com/Zhaoyang-Ma/FedTopo.

Problem

• federated learning
• model heterogeneity
• representation alignment
• knowledge sharing
• class topology

Innovation

• relation-level topology
• model-heterogeneous federated learning
• class relation topology
• reliability-aware aggregation
• feature space alignment