Research map / Federated learning and privacy
Differential privacy: research map
838 papers on Differential privacy in Federated learning and privacy, from ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR (2016–2026), grouped into 2 clusters and 9 approaches. The busiest year so far is 2026.
What it is
Differential privacy is a mathematical guarantee that a model or statistic reveals almost nothing about any single person in the data. This topic covers private training (DP-SGD), private statistics and synthetic data, and the accounting of how much privacy is spent.
Terms you will see on this page:
- Differential privacy (DP)
- a guarantee that outputs barely change if one person's data is added or removed.
- DP-SGD
- SGD with clipped per-example gradients plus noise.
- Privacy budget (epsilon)
- how much privacy loss is allowed; smaller is more private.
Start reading with the most cited papers here:
- Learning Differentially Private Recurrent Language Models (ICLR 2018 · 743 citations)
- Scalable Private Learning with PATE (ICLR 2018 · 313 citations)
Within Federated learning and privacy, its share held steady from 28.9% in 2023–24 to 29.1% in 2025–26 (215 → 266 papers at ICML, NeurIPS, ICLR, CVPR and AAAI, the venues with data for all four years).
Explore Differential privacy in the interactive map
Working on something in this topic? Describe your idea in scime atlas to see which approach it falls under, the closest papers by meaning and how crowded the spot has become.
What changed recently
Approaches whose share of the topic moved most between 2023–24 and 2025–26 (ICML, NeurIPS, ICLR, CVPR and AAAI):
- Differentially private synthetic data rising: 8% → 18% of the topic (17 → 48 papers)
- Differentially private federated learning rising: 2% → 9% of the topic (5 → 23 papers)
- Privacy accounting and mechanisms rising: 14% → 16% of the topic (30 → 42 papers)
- Private training with public data falling: 19% → 13% of the topic (40 → 34 papers)
- Private data analysis frameworks falling: 13% → 9% of the topic (28 → 23 papers)
Papers from 2025–2026 that are already cited most:
- Enabling Differentially Private Federated Learning for Speech Recognition: Benchmarks, Adaptive Optimizers, and Gradient Clipping (NeurIPS 2025 · 5 citations)
- Federated t-SNE and UMAP for Distributed Data Visualization (AAAI 2025 · 5 citations)
- Clients Collaborate: Flexible Differentially Private Federated Learning with Guaranteed Improvement of Utility-Privacy Trade-off (ICML 2025 · 3 citations)
- Sum Estimation under Personalized Local Differential Privacy (NeurIPS 2025 · 1 citations)
Approaches and key papers
Private estimation and analysis · 546 papers
Keywords: estimation, error, algorithms
Approaches in this cluster:
- Privacy accounting and mechanisms (135 papers)
Develop tighter composition bounds and noise mechanisms for differential privacy. - Private online and user-level learning (128 papers)
Analyze private learning in online, shuffle and user-level settings. - Local-DP and private estimation (125 papers)
Estimate means and statistics under local or central privacy with minimax rates. - Private data analysis frameworks (116 papers)
Design private release, counting and sensitivity-bounding methods for data analysis. - Differentially private clustering (42 papers)
Give approximation algorithms for private k-means and clustering.
Most cited and most cited since 2024:
- Scalable Private Learning with PATE (ICLR 2018 · 313 citations)
- Collecting Telemetry Data Privately (NeurIPS 2017 · 262 citations)
- Prior-itizing Privacy: A Bayesian Approach to Setting the Privacy Budget in Differential Privacy (NeurIPS 2024 · 9 citations)
- One-shot Empirical Privacy Estimation for Federated Learning (ICLR 2024 · 5 citations)
Private model training · 292 papers
Keywords: dp sgd, gradient, clipping
Approaches in this cluster:
- Private training with public data (101 papers)
Reduce accuracy loss of differentially private models with public data, finetuning and new mechanisms. - Differentially private synthetic data (87 papers)
Generate private synthetic data or text and in-context examples under differential privacy. - DP-SGD refinements (61 papers)
Make private stochastic gradient descent cheaper and more practical, without sampling or at scale. - Differentially private federated learning (43 papers)
Apply clipping and noise in federated learning with client-level privacy.
Most cited and most cited since 2024:
- Learning Differentially Private Recurrent Language Models (ICLR 2018 · 743 citations)
- PATE-GAN: Generating Synthetic Data with Differential Privacy Guarantees (ICLR 2019 · 278 citations)
- DP-AdamBC: Your DP-Adam Is Actually DP-SGD (Unless You Apply Bias Correction) (AAAI 2024 · 12 citations)
- Position: Considerations for Differentially Private Learning with Large-Scale Public Pretraining (ICML 2024 · 9 citations)
