Research map / Multi-agent systems and games / Cooperative multi-agent RL
A Multi-Agent Reinforcement Learning Approach for Efficient Client Selection in Federated Learning
Sai Qian Zhang, Jieyu Lin, Qi Zhang
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use MARL for runtime client selection based on probing loss and system statistics | Federated learning · MNIST, CIFAR-10, Fashion MNIST, Shakespeare | test accuracy: 96.91% vs 96.86%; 48.87% vs 48.19%; 96.14% vs 96.08%; 44.58% vs 44.10% | Improved |
| Use probing loss to select clients, rejecting those with above-average probing loss | FedAvg under non-IID client data · MNIST and CIFAR-10 | test accuracy: 98.7% and 45.2%; 1.3% and 2.8% higher than the other methods on average | Improved |
| Use multi-agent reinforcement learning rather than single-agent reinforcement learning | FedMarl · Federated learning training environment | convergence speed | Improved |
| Increase latency and communication-cost reward weights from 0.2 and 0.1 to 0.5 and 0.3 (ablation) | FedMarl · LeNet, VGG6, ResNet-18 | processing latency: 1.0× -> 0.82×; 1.0× -> 0.80×; 1.0× -> 0.80× | Improved |
| Increase latency and communication-cost reward weights from 0.2 and 0.1 to 0.5 and 0.3 (ablation) | FedMarl · LeNet, VGG6, ResNet-18 | communication cost: 1.0× -> 0.83×; 1.0× -> 0.78×; 1.0× -> 0.86× | Improved |
| Increase latency and communication-cost reward weights from 0.2 and 0.1 to 0.5 and 0.3 (ablation) | FedMarl · LeNet, VGG6, ResNet-18 | test accuracy: 98.15% -> 97.88%; 46.05% -> 45.28%; 98.43% -> 98.03% | Hurt |
| Use agents trained on another DNN and dataset without retraining | FedMarl · LeNet on MNIST | normalized reward: 0.965 (0.989) | Mixed |
| Use normalized agent-state inputs to transfer trained agents to different client-pool sizes | FedMarl · CIFAR-10, K=400 and K=800 | total reward: 1.36× and 1.32× higher ... on average | Improved |
| Use normalized agent-state inputs to transfer trained agents to a pool with new device types | FedMarl · CIFAR-10 with six additional mobile device types | overall performance | Improved |
| Reduce processing latency and communication cost through FedMarl client selection | Federated learning · Evaluation benchmarks | processing latency: 1.7× | Improved |
| Reduce processing latency and communication cost through FedMarl client selection | Federated learning · Evaluation benchmarks | communication cost: 2.2× | Improved |
Closest papers by meaning
- Finite-Time Analysis of On-Policy Heterogeneous Federated Reinforcement Learning (ICLR 2024 · similarity 0.90)
- Federated Offline Reinforcement Learning: Collaborative Single-Policy Coverage Suffices (ICML 2024 · similarity 0.89)
- Federated Ensemble-Directed Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
- Federated Q-Learning: Linear Regret Speedup with Low Communication Cost (ICLR 2024 · similarity 0.88)
- Momentum for the Win: Collaborative Federated Reinforcement Learning across Heterogeneous Environments (ICML 2024 · similarity 0.88)
- Reinforcement Active Client Selection for Federated Heterogeneous Graph Learning (AAAI 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
