Conference

Byzantine-Resilient Federated Policy Gradient for Reinforcement Learning

Wen-Yuan Tian Jing-Xuan Zhou Miao Yu Xin-Fa Huang

Aug 2026 · 2026 12th International Conference on Big Data and Information Analytics (BigDIA) · pp. 896-901 · 0 citations · 20 references

Abstract

Federated reinforcement learning (FedRL) permits agents to optimize a shared policy without exchanging raw trajectories, but noisy policy-gradient estimates make malicious updates difficult to distinguish from honest variation. We propose FedPG-BR, which couples stochastic variance-reduced policy gradient (SVRPG) updates with a dual-filtering aggregator. Variance reduction concentrates honest updates, while median-centered filtering removes large deviations and retains a set of central gradients for averaging. Under smoothness, bounded-variance and bounded-importance-weight assumptions, FedPG-BR retains the $\mathcal{O}\left( {{\varepsilon ^{ - 5/3}}} \right)$ benign sample-complexity order and admits a per-round Byzantine perturbation term when the adversarial fraction is below 0.5. Experiments on three OpenAI Gym tasks under three static attacks indicate improved robustness over FedAvg-PG, Krum and coordinate-wise median baselines. The aggregator requires no additional communication round and transmits the same order of gradient values as these baselines. The current evidence is limited to centralized synthetic simulations; dynamic participation and multi-robot deployment remain to be evaluated.

DOI Semantic Scholar

View source

Similar papers

2026

FedDPRL: Communication-Efficient Differentially Private Federated Reinforcement Learning With Adaptive Gradient Compression

Cross-silo federated reinforcement learning (FRL) trains a shared policy across silos that must simultaneously respect a per-round uplink budget and on-device privacy. Although differential privacy (DP) and gradient compression each have mature solutions in supervised federated learning, naively stacking them in the po...

Jian-Tao Xu · IEEE Transactions on Communi... · 0 citations

#artificial intelligence Preprint Aug 2026

Provably Efficient Federated Reinforcement Learning with Linear Function Approximation and Logarithmic Communication Cost

Fed-LSVI is proposed, the first provably efficient federated algorithm for online reinforcement learning with linear function approximation in episodic Markov decision processes and achieves a regret bound of $\widetilde{\mathcal O}(\sqrt{Md^3H^4T})$, matching the best-known regret for multi-agent online reinforcement...

Zi-Han Liang, Haochen Zhang, Ling-Zhou Xue · 0 citations

Open access Aug 2026

InfRA-FL: Information-Driven Robust Federated Learning via Saturation-Aware Reinforcement Learning

Experiments show that InfRA-FL outperforms state-of-the-art baselines by 5–10% in accuracy while accelerating convergence, establishing that principled information-theoretic observation and robust reward design suffice to secure RL-driven federated learning against targeted poisoning.

Jiao Tian, Jinlin He, Lie-Jun Wang · Machine Learning and Knowled... · 0 citations

#machine learning Preprint Sep 2026

Robust Decentralized Personalized Federated Learning via Prediction-Constrained Neighborhood Collaboration

This paper proposes a robust decentralized personalized federated learning method R-DPFL, that enables clients to reduce the impact of Byzantine attacks via robust neighborhood direction estimation and history-based update trend prediction, rather than purely aggregating client models as in the existing work. In R-DPFL...

Xiao Ma, Hong Shen, Hui Tian et al. · 1 citation

#machine learning Preprint Sep 2026

Fully Byzantine-Resilient Multi-Agent Reinforcement Learning

It is proved that agents'parameters converge almost surely to the same limit points as in the attack-free case over time-varying communication graphs under linear parameterizations of the value and team-reward functions and Byzantine edge attacks.

Haejoon Lee, Dimitra Panagou · 0 citations

#machine learning Preprint Sep 2026

FedeRage: Provably Convergent Agnostic Federated Learning under General Client Drift

Federated learning (FL) enables collaborative model training without sharing raw data, but its performance degrades under non-IID data and stochastic client participation. Remedies built on classical Federated Averaging (FedAvg) typically presuppose that client participation probabilities are known to the server, which...

Herlock Rahimi, Dionysis Kalogerias · 0 citations