Research map / Federated learning and privacy / Personalized and heterogeneous FL

Reinforcement Active Client Selection for Federated Heterogeneous Graph Learning

AAAI 2025 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Federated semi-supervised and graph learning · 6 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes RAFHGL, a reinforcement-learning-based client selection method for federated heterogeneous graph learning that combines an active learning agent on clients, a DDQN-trained client selection agent on the server, and prototype-based local loss correction.

Problem

Client selection in federated heterogeneous graph learning (FHGL) is hard because heterogeneous graph data make client contributions difficult to assess and data heterogeneity creates a trade-off between convergence and accuracy.

Approach

Combines a local DQN-based active learning agent that selects informative nodes and reports statistics of active scores with a global DDQN client selection agent that uses these statistics as state, and mitigates the resulting client drift with data-prototype regularization.

Key results

• RAFHGL reaches 56.62% on Academic (uniform) and 69.01% (Dirichlet), outperforming the best baseline in most settings per Table 2.
• On DBLP ablation, RAFHGL reaches 75.19% accuracy in 52 rounds versus 95 rounds for the prototype-only variant P.
• With 2 of 10 clients selected on Academic, RAFHGL converges in 58 rounds, about 36% fewer than Random's 91.

Measured effects reported in the paper

Surprising findings

• Some baselines, such as FedMarl and FedCor, perform worse than random selection on several datasets, e.g. FedCor reaches 69.35% on DBLP Dirichlet and 66.88% on Yelp Dirichlet against 74.84% for Random.

Limitations

• Relies on hand-tuned hyperparameters such as the strategy probability epsilon and the quantile n_q, which strongly affect results.
• Edges are discarded when partitioning graphs, so partitioning effects are confounded with data heterogeneity.

Experimental setup

Datasets: Academic DBLP ACM Yelp

Benchmarks: Academic DBLP ACM Yelp

Metrics: accuracy convergence rounds training loss

Baselines: Random Cluster-N Cluster-G DivFL FedCor pow-d AFL ACFL FedMarl FAVOR

Builds on: FAVOR DDQN DQN HAN FedMarl

Closest papers by meaning

• Heterogeneity-Guided Client Sampling: Towards Fast and Efficient Non-IID Federated Learning (NeurIPS 2024 · similarity 0.89)
• Exact and Linear Convergence for Federated Learning under Arbitrary Client Participation is Attainable (NeurIPS 2025 · similarity 0.86)
• FedRSPO+: A Heterogeneity-aware Algorithm for Decision-focused Federated Learning (NeurIPS 2026 · similarity 0.86)
• Finite-Time Analysis of On-Policy Heterogeneous Federated Reinforcement Learning (ICLR 2024 · similarity 0.85)
• Momentum for the Win: Collaborative Federated Reinforcement Learning across Heterogeneous Environments (ICML 2024 · similarity 0.85)
• Convergence Analysis of Split Federated Learning on Heterogeneous Data (NeurIPS 2024 · similarity 0.85)