Research map / Federated learning and privacy / Communication-efficient FL
Local drift correction
Regularize or reaggregate local training to counter heterogeneity and forgetting in federated learning.
61 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.
24 papers in 2023–24 and 16 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 16.2% to 13.8%.
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 · Personalized Federated Learning with Moreau Envelopes (NeurIPS 2020 · 213 citations)
- 2021 · Adaptive Federated Optimization (ICLR 2021 · 122 citations)
- 2022 · FedAvg with Fine Tuning: Local Updates Lead to Representation Learning (NeurIPS 2022 · 46 citations)
- 2023 · Elastic Aggregation for Federated Optimization (CVPR 2023 · 49 citations)
- 2024 · Efficient Asynchronous Federated Learning with Prospective Momentum Aggregation and Fine-Grained Correction (AAAI 2024 · 31 citations)
- 2025 · Creating Coherence in Federated Non-Negative Matrix Factorization (AAAI 2025 · 0 citations)
- 2026 · Federated Bilevel Performative Prediction (ICML 2026 · 0 citations)
Earliest papers here
- Personalized Federated Learning with Moreau Envelopes (NeurIPS 2020 · 213 citations)
- Adaptive Federated Optimization (ICLR 2021 · 122 citations)
- Game of Gradients: Mitigating Irrelevant Clients in Federated Learning (AAAI 2021 · 85 citations)
What has been tried
50 papers here report 420 measured changes (a component added, removed or swapped, compared with a baseline): helped 73%, no clear effect 11%, hurt 12%, 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 |
|---|---|---|---|---|
| Add client weight evaluation based on temporal gradient | CIFAR-10 · accuracy | Helped | Efficient Asynchronous Federated Learning with Prospective Momentum Aggregation and Fine-Grained Correction (AAAI 2024 · 31 citations) | |
| add partial personalization to DFedAvgM | CIFAR-10 Dirichlet α=0.3 · test accuracy (%) | 82.49 -> 84.69 | Helped | Decentralized Directed Collaboration for Personalized Federated Learning (CVPR 2024 · 31 citations) |
| increase cut layer L_c so clients maintain a larger portion of the global model | CIFAR-10 and CIFAR-100 · accuracy | Helped | Convergence Analysis of Split Federated Learning on Heterogeneous Data (NeurIPS 2024 · 15 citations) | |
| Vary the noise scale | consensus problem · objective value | Mixed | z-SignFedAvg: A Unified Stochastic Sign-Based Compression for Federated Learning (AAAI 2024 · 12 citations) | |
| Add client-group correction to HFedAvg | HFL with client non-i.i.d. within groups · testing accuracy | Helped | Hierarchical Federated Learning with Multi-Timescale Gradient Correction (NeurIPS 2024 · 11 citations) | |
| Replace local SGD with SAM | CIFAR-10 and CIFAR-100, IID/Dir-1.0/Dir-0.1, LeNet and ResNet-18 · test accuracy | 85.31 -> 86.47; 84.94 -> 85.90; 80.28 -> 81.96; 51.41 -> 55.56; 51.17 -> 54.62; 48.15 -> 53.15; 85.11 -> 87.44; 84.33 -> | Helped | A-FedPD: Aligning Dual-Drift is All Federated Primal-Dual Learning Needs (NeurIPS 2024 · 4 citations) |
| Add the FedProx proximal loss to Δ-SGD | CIFAR-10 ResNet-18, α=0.01 · accuracy | Mixed | Adaptive Federated Learning with Auto-Tuned Clients (ICLR 2024 · 3 citations) | |
| Use momentum coefficient β=1, turning off momentum | Full client participation · convergence rate | Hurt | Momentum Benefits Non-iid Federated Learning Simply and Provably (ICLR 2024 · 2 citations) | |
| set delay threshold to $\tau_c=8$ rather than $\tau_c=1$, $4$, or $10$ | CIFAR-10, large worst-case delay · accuracy | Helped | FADAS: Towards Federated Adaptive Asynchronous Optimization (ICML 2024 · 1 citation) | |
| add personalized alignment (LAPρ) | Fashion MNIST · RMSD | 7.734 -> 6.733 | Helped | Creating Coherence in Federated Non-Negative Matrix Factorization (AAAI 2025 · 0 citations) |
| Replace cosine similarity with another similarity measure | CIFAR-10 x=0.1, 0.5, 1 average · accuracy | 74.14; 75.69; 76.56 | Mixed | FedQS: Optimizing Gradient and Model Aggregation for Semi-Asynchronous Federated Learning (NeurIPS 2025 · 0 citations) |
| lower activation probability in DLR-Random | i-MNIST non-stationary · Accum. Error Rate | 0.0553 -> 0.0622 | Hurt | Event-Driven Online Vertical Federated Learning (ICLR 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 11 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 · 5 of 11 papers
- MNIST · 5 of 11 papers
- Fashion MNIST · 2 of 11 papers
Baselines
- FedAvg · 5 of 11 papers
- FedProx · 2 of 11 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 40 papers; from OpenAlex and the papers' affiliations).
- JD Explore Academy · 5 papers
- The Chinese University of Hong Kong · 3 papers
- IBM · 3 papers
- The University of Sydney · 3 papers
- George Mason University · 2 papers
- Rice University · 2 papers
- Tsinghua University · 2 papers
- Shanghai Jiao Tong University · 2 papers
Use it
Download BibTeX of 35 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)
- Federated optimization algorithms (70 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).