Research map / Federated learning and privacy / Personalized and heterogeneous FL
FedECADO: A Dynamical System Model of Federated Learning
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 FedECADO, a federated learning algorithm derived from a circuit/dynamical-system view of ECADO, adding an aggregate sensitivity model scaled by client data share and a multi-rate backward-Euler integration with adaptive step sizes to handle non-IID data and heterogeneous client computation.
Problem
Federated learning suffers from inconsistent updates when client data distributions are non-IID and client computation (learning rates, epochs) is heterogeneous, limiting global model accuracy.
Approach
Model federated training as a continuous-time circuit/ODE system (extending ECADO): a dataset-size-weighted linear sensitivity term per client couples the central state, and a multi-rate integration with linear interpolation/extrapolation of client states plus LTE-based adaptive step sizes synchronizes asynchronous client updates in continuous time.
Key results
- On non-IID CIFAR-10 with VGG-11 (100 clients, Dir(0.1)), FedECADO reached 57.8% mean accuracy vs 48.9% for FedNova and 44.3% for FedProx over 20 partitions.
- With heterogeneous learning rates and epochs on CIFAR-10, FedECADO achieved 72.2% mean accuracy vs 67.6% (FedNova) and 61.4% (FedProx).
- On ResNet-34 with CIFAR-100 under non-IID and asynchronous settings, FedECADO was 4.6% higher than FedNova and 8.6% higher than FedProx in accuracy after 100 epochs.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add an aggregate sensitivity model scaled by each client's relative dataset size | FedECADO · CIFAR-10, non-IID Dirichlet, 100 clients, 200 epochs | classification accuracy (%): 48.9 -> 57.8 | Helped |
| add multi-rate integration to synchronize heterogeneous client updates | FedECADO · CIFAR-10, IID, heterogeneous client learning rates and epochs, 100 clients, 100 epochs | classification accuracy (%): 67.6 -> 72.2 | Helped |
| combine aggregate sensitivity and multi-rate integration | FedECADO · CIFAR-100, non-IID and heterogeneous computation, 100 clients, 100 epochs | classification accuracy (%): +4.6% vs FedNova | Helped |
Limitations
- Convergence proof relies on strong assumptions (bounded, coercive, Lipschitz gradients) and contraction argument that is not fully rigorous.
- Sensitivity uses a constant precomputed averaged Hessian rather than per-step Hessians.
- Experiments limited to image classification with VGG-11 and ResNet-34 and only two baselines.
Experimental setup
Datasets: CIFAR-10 CIFAR-100
Benchmarks: CIFAR-10 CIFAR-100
Metrics: classification accuracy training loss
Baselines: FedProx FedNova
Builds on: ECADO FedAvg FedProx FedNova
Closest papers by meaning
- FedSMU: Communication-Efficient and Generalization-Enhanced Federated Learning through Symbolic Model Updates (ICML 2025 · similarity 0.89)
- Adaptive Federated Learning with Auto-Tuned Clients (ICLR 2024 · similarity 0.89)
- Efficient Federated Learning against Heterogeneous and Non-stationary Client Unavailability (NeurIPS 2024 · similarity 0.88)
- Exact and Linear Convergence for Federated Learning under Arbitrary Client Participation is Attainable (NeurIPS 2025 · similarity 0.88)
- Efficient Adaptive Federated Optimization (NeurIPS 2025 · similarity 0.88)
- FADAS: Towards Federated Adaptive Asynchronous Optimization (ICML 2024 · 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.