Research map / Federated learning and privacy / Communication-efficient FL
Federated optimization algorithms
Develop faster local-update and composite or minimax optimizers for federated learning.
70 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2020–2026) · an approach in Communication-efficient FL, keyword cluster “optimization · convex · fedavg”
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.
30 papers in 2023–24 and 20 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 20.3% to 17.2%.
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
- 2020 · SCAFFOLD: Stochastic Controlled Averaging for Federated Learning (ICML 2020 · 709 citations)
- 2021 · On the Convergence of Communication-Efficient Local SGD for Federated Learning (AAAI 2021 · 59 citations)
- 2022 · FedNL: Making Newton-Type Methods Applicable to Federated Learning (ICML 2022 · 12 citations)
- 2023 · Single-shot General Hyper-parameter Optimization for Federated Learning (ICLR 2023 · 13 citations)
- 2024 · FedNS: A Fast Sketching Newton-Type Algorithm for Federated Learning (AAAI 2024 · 4 citations)
- 2025 · Compressed and distributed least-squares regression: convergence rates with applications to federated learning (ICML 2025 · 104 citations)
- 2026 · FedAdamW: A Communication-Efficient Optimizer with Convergence and Generalization Guarantees for Federated Large Models (AAAI 2026 · 1 citation)
Earliest papers here
- SCAFFOLD: Stochastic Controlled Averaging for Federated Learning (ICML 2020 · 709 citations)
- Lower Bounds and Optimal Algorithms for Personalized Federated Learning (NeurIPS 2020 · 89 citations)
- FetchSGD: Communication-Efficient Federated Learning with Sketching (ICML 2020 · 65 citations)
What has been tried
61 papers here report 496 measured changes (a component added, removed or swapped, compared with a baseline): helped 66%, no clear effect 11%, hurt 18%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Learn projection A instead of keeping it frozen | MNLI, 50 clients · accuracy | Helped | Robust Federated Finetuning of LLMs via Alternating Optimization of LoRA (NeurIPS 2025 · 4 citations) | |
| Increase the sketch size | LIBSVM datasets: phishig, cod-rna, covtype, and SUSY · generalization performance | Helped | FedNS: A Fast Sketching Newton-Type Algorithm for Federated Learning (AAAI 2024 · 4 citations) | |
| Increase the stochastic-gradient batch size | kPCA on MNIST · convergence speed | Helped | Nonconvex Federated Learning on Compact Smooth Submanifolds With Heterogeneous Data (NeurIPS 2024 · 3 citations) | |
| Vary the communication interval | Camelyon17 · performance | No clear effect | Communication-Efficient Federated Group Distributionally Robust Optimization (NeurIPS 2024 · 3 citations) | |
| Use β=0.2 rather than other tested momentum values | MNIST and Fashion MNIST · test accuracy | Helped | Stochastic Controlled Averaging for Federated Learning with Communication Compression (ICLR 2024 · 1 citation) | |
| replace concatenation with summarization | BBH Object Counting · accuracy | 0.90 -> 0.88 | Hurt | Can Textual Gradient Work in Federated Learning? (ICLR 2025 · 0 citations) |
| enable forced synchronization | Tabular episodic MDPs · regret | No clear effect | Federated $Q$-Learning with Reference-Advantage Decomposition: Almost Optimal Regret and Logarithmic Communication Cost (ICLR 2025 · 0 citations) | |
| apply clipping to the final pseudogradient once per epoch instead of clipping at every step | Synthetic 1D function · function residual | Helped | Methods with Local Steps and Random Reshuffling for Generally Smooth Non-Convex Federated Optimization (ICLR 2025 · 0 citations) | |
| Construct the mapping from a different dataset rather than the target dataset | CIFAR10, target accuracy 78% · communication bits | 0.0068 -> 0.0075 GiB | No clear effect | Achieving Lossless Gradient Sparsification via Mapping to Alternative Space in Federated Learning (ICML 2024 · 0 citations) |
| Upgrade more graph-data agents from SRW to SRRW | ijcnn1 binary classification · asymptotic MSE | Helped | Does Worst-Performing Agent Lead the Pack? Analyzing Agent Dynamics in Unified Distributed SGD (NeurIPS 2024 · 0 citations) | |
| Add full-precision momentum completion | CIFAR-100, Dir (0.25), LeNet · Top-1 test accuracy (%) | 52.35 -> 52.53 | Helped | FedSMU: Communication-Efficient and Generalization-Enhanced Federated Learning through Symbolic Model Updates (ICML 2025) |
| use multi-step inner updates (K > 1) instead of the one-step update (K = 1) | CATvsDOG and CIFAR10 · AUC | Helped | A Federated Stochastic Multi-level Compositional Minimax Algorithm for Deep AUC Maximization (ICML 2024) |
What reviewers will expect
Datasets and baselines named in the 14 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 · 3 of 14 papers
Baselines
- EF21 · 2 of 14 papers
- FedAvg · 2 of 14 papers
- SCAFFOLD · 2 of 14 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 50 papers; from OpenAlex and the papers' affiliations).
- Carnegie Mellon University · 4 papers
- Purdue University · 3 papers
- Pennsylvania State University · 3 papers
- University of Chicago · 3 papers
- State University of New York at Buffalo · 3 papers
- LinkedIn · 2 papers
- Texas A&M University - College Station · 2 papers
- Technion - Israel Institute of Technology · 2 papers
- École Polytechnique · 2 papers
- Ecole Polytechnique · 2 papers
Use it
Download BibTeX of 38 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
- Client participation analysis (68 papers)
- Local drift correction (61 papers)
Elsewhere in the topic
- Efficient local training in FL (77 papers)
- Communication-efficient heterogeneous FL (65 papers)
- Decentralized peer-to-peer training (41 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).