Research map / Federated learning and privacy / Personalized and heterogeneous FL
FedRolex: Model-Heterogeneous Federated Learning with Rolling Sub-Model Extraction
Samiul Alam, Luyang Liu, Ming Yan, Mi Zhang
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| rolling sub-model extraction instead of static (HeteroFL) | partial training FL · CIFAR-10 high heterogeneity | global accuracy: 63.90 -> 69.44 | Improved |
| rolling sub-model extraction instead of random (Federated Dropout) | partial training FL · CIFAR-10 high heterogeneity | global accuracy: 46.64 -> 69.44 | Improved |
| rolling extraction vs Federated Dropout | partial training FL · CIFAR-10 low heterogeneity | global accuracy: 76.20 -> 84.45 | Improved |
| rolling extraction vs HeteroFL | partial training FL · Stack Overflow | global accuracy: 27.21 -> 29.22 | Improved |
| FedRolex vs KD-based Fed-ET with public data | FedRolex · CIFAR-10 high heterogeneity | global accuracy: 78.66 -> 69.44 | Hurt |
| FedRolex vs Fed-ET without public data | FedRolex · CIFAR-100 high heterogeneity | global accuracy: 35.78 -> 56.57 | Improved |
| model-heterogeneous FedRolex vs homogeneous upper bound | FedRolex · Stack Overflow | global accuracy: 29.79 -> 29.55 | No clear effect |
| add client weighting schemes to selective averaging (ablation) | FedRolex · CIFAR-10 high heterogeneity | global accuracy: 69.44 -> 69.09 | No clear effect |
| vary client participation rate from 5% to 20% | FedRolex · CIFAR-10 high heterogeneity | global accuracy: 57.90 -> 71.85 | Improved |
| vary overlap between rolling windows (ablation) | FedRolex · CIFAR-10/100 | global accuracy | No clear effect |
| increase global model size to 2-16x client model (ablation) | FedRolex · CIFAR-10/100 | global accuracy | Hurt |
| increase fraction of large-capacity clients beyond a small share (ablation) | FedRolex · CIFAR/Stack Overflow | global accuracy | No clear effect |
Later papers that compared with this method
6 later accepted papers report FedRolex as a baseline in their results tables; 6 of them report a better number on most of the results they share with it.
- Breaking the Capacity Bottleneck in Model-Heterogeneous Federated Learning via Gradual Model Restoration (ICML 2026 · better on 4 of 4 shared results)
- FedLAGC: Towards High Performance System-Heterogeneous Federated Learning via Layer-Adaptive Submodel Extraction and Gradient Correction (AAAI 2026 · better on 2 of 2 shared results)
- Cluster Based Heterogeneous Federated Foundation Model Adaptation and Fine-Tuning (AAAI 2025 · better on 4 of 7 shared results)
- FedLMT: Tackling System Heterogeneity of Federated Learning via Low-Rank Model Training with Theoretical Guarantees (ICML 2024 · better on 1 of 1 shared result)
- Aggregating Capacity in FL through Successive Layer Training for Computationally-Constrained Devices (NeurIPS 2023 · better on 8 of 8 shared results)
- Resolving the Tug-of-War: A Separation of Communication and Learning in Federated Learning (NeurIPS 2023 · better on 4 of 4 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- One-Shot Heterogeneous Federated Learning with Local Model-Guided Diffusion Models (ICML 2025 · similarity 0.91)
- FedEL: Federated Elastic Learning for Heterogeneous Devices (NeurIPS 2025 · similarity 0.89)
- Federated Model Heterogeneous Matryoshka Representation Learning (NeurIPS 2024 · similarity 0.89)
- FedGPS: Statistical Rectification Against Data Heterogeneity in Federated Learning (NeurIPS 2025 · similarity 0.89)
- Federated Learning with Heterogeneous LLMs: Integrating Small Student Client Models with a Large Hungry Model (AAAI 2025 · similarity 0.89)
- FIARSE: Model-Heterogeneous Federated Learning via Importance-Aware Submodel Extraction (NeurIPS 2024 · similarity 0.89)
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.
