Research map / Federated learning and privacy / Communication-efficient FL
Efficient local training in FL
Cut federated training cost with low precision, few rounds and decoupled local learning.
77 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2018–2026) · an approach in Communication-efficient FL, keyword cluster “devices · edge · dynamic”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
29 papers in 2023–24 and 31 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 19.6% to 26.7%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2018 · Distributed Fine-tuning of Language Models on Private Data (ICLR 2018 · 11 citations)
- 2020 · Group Knowledge Transfer: Federated Learning of Large CNNs at the Edge (NeurIPS 2020 · 187 citations)
- 2021 · HeteroFL: Computation and Communication Efficient Federated Learning for Heterogeneous Clients (ICLR 2021 · 192 citations)
- 2022 · SplitFed: When Federated Learning Meets Split Learning (AAAI 2022 · 696 citations)
- 2023 · FedDM: Iterative Distribution Matching for Communication-Efficient Federated Learning (CVPR 2023 · 103 citations)
- 2024 · FedASMU: Efficient Asynchronous Federated Learning with Dynamic Staleness-Aware Model Update (AAAI 2024 · 73 citations)
- 2025 · Decentralized Federated Learning with Model Caching on Mobile Agents (AAAI 2025 · 6 citations)
- 2026 · FedCure: Mitigating Participation Bias in Semi-Asynchronous Federated Learning with Non-IID Data (AAAI 2026 · 1 citation)
Earliest papers here
- Distributed Fine-tuning of Language Models on Private Data (ICLR 2018 · 11 citations)
- Group Knowledge Transfer: Federated Learning of Large CNNs at the Edge (NeurIPS 2020 · 187 citations)
- Federated Bayesian Optimization via Thompson Sampling (NeurIPS 2020)
What has been tried
61 papers here report 645 measured changes (a component added, removed or swapped, compared with a baseline): helped 65%, no clear effect 8%, hurt 22%, mixed 5%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add dynamic model aggregation on the server | CIFAR-10 · accuracy | 1.38\% to 4.32\% | Helped | FedASMU: Efficient Asynchronous Federated Learning with Dynamic Staleness-Aware Model Update (AAAI 2024 · 73 citations) |
| Add budget-aware extrusion | CIFAR-10, ResNet18 · accuracy | Helped | FedMef: Towards Memory-efficient Federated Dynamic Pruning (CVPR 2024 · 20 citations) | |
| Increase shared encoder depth from ⌈0.25N⌉ to ⌈0.5N⌉ | IID average · accuracy (%) | 86.27 -> 82.04 | Hurt | FedLPS: Heterogeneous Federated Learning for Multiple Tasks with Local Parameter Sharing (AAAI 2024 · 9 citations) |
| Increase embedding dimension | FEMNIST, C=8 · F1 | 6.1 -> 67.4 | Helped | TurboSVM-FL: Boosting Federated Learning through SVM Aggregation for Lazy Clients (AAAI 2024 · 9 citations) |
| increase the number of devices from N=10 to N=50 | CIFAR-100, ResNet-18 · accuracy | 68.31% -> 67.58% | No clear effect | Device-Wise Federated Network Pruning (CVPR 2024 · 8 citations) |
| Add non-uniform seed probabilities based on scalar-gradient magnitude | Natural Instructions and Dolly-15K test scenarios · Rouge-L | 22.33 -> 23.50; 29.77 -> 30.19; 32.90 -> 33.18; 35.64 -> 36.29; 33.12 -> 33.00; 35.93 -> 35.95 | Mixed | Federated Full-Parameter Tuning of Billion-Sized Language Models with Communication Cost under 18 Kilobytes (ICML 2024 · 7 citations) |
| increase cache size from 1 to 30 | MNIST, FashionMNIST non-iid · test accuracy | Helped | Decentralized Federated Learning with Model Caching on Mobile Agents (AAAI 2025 · 6 citations) | |
| add generative activations to the server-side update | Fashion-MNIST, shard=2 · gradient dissimilarity | Helped | GAS: Generative Activation-Aided Asynchronous Split Federated Learning (AAAI 2025 · 6 citations) | |
| Add RI restoration to regularized AFL | CIFAR-100, K=1000, γ=100 · top-1 accuracy (%) | 49.62 -> 58.56 | Helped | AFL: A Single-Round Analytic Approach for Federated Learning with Pre-trained Models (CVPR 2025 · 6 citations) |
| Add FedPop-G (inter-configuration tuning) to RS | CIFAR-10 IID / NIID Dir1.0 / NIID Dir0.5 · accuracy (global) | +1.57 IID, +3.34 Dir1.0, +2.75 Dir0.5 | Helped | FedPop: Federated Population-based Hyperparameter Tuning (AAAI 2025 · 5 citations) |
| add early stopping | EMNIST · accuracy (%) | 0.11% | No clear effect | FedSPU: Personalized Federated Learning for Resource-Constrained Devices with Stochastic Parameter Update (AAAI 2025 · 5 citations) |
| distill the Fens ensemble into a single ResNet-8 | CIFAR-10 α=0.01, 0.05, and 0.1 · accuracy | 44.20 -> 43.81; 68.22 -> 65.56; 75.61 -> 71.59 | Hurt | Revisiting Ensembling in One-Shot Federated Learning (NeurIPS 2024 · 4 citations) |
What reviewers will expect
Datasets and baselines named in the 20 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CIFAR-10 · 14 of 20 papers
- MNIST · 6 of 20 papers
- CIFAR-100 · 5 of 20 papers
- SVHN · 4 of 20 papers
- Fashion-MNIST · 3 of 20 papers
- Tiny-ImageNet · 3 of 20 papers
- FEMNIST · 2 of 20 papers
- ImageNet-1K · 2 of 20 papers
Baselines
- FedAvg · 9 of 20 papers
- FedProx · 3 of 20 papers
- FedDST · 2 of 20 papers
- FedTiny · 2 of 20 papers
- FedMef · 2 of 20 papers
- FLASH · 2 of 20 papers
- Moon · 2 of 20 papers
- FedGen · 2 of 20 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 60 papers; from OpenAlex and the papers' affiliations).
- University of Pittsburgh · 3 papers
- Hong Kong University of Science and Technology · 3 papers
- Meta · 3 papers
- North Carolina State University · 2 papers
- Technical University of Munich · 2 papers
- Massachusetts Institute of Technology · 2 papers
- Wuhan University of Science and Technology · 2 papers
- Zhejiang Normal University · 2 papers
- Key Laboratory of Social Computing and Cognitive Intelligence (Dalian University of Technology) · 2 papers
- South China University of Technology · 2 papers
Use it
Download BibTeX of 37 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Communication-efficient heterogeneous FL (65 papers)
- Decentralized peer-to-peer training (41 papers)
Elsewhere in the topic
- Federated optimization algorithms (70 papers)
- Client participation analysis (68 papers)
- Local drift correction (61 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).