Research map / Learning theory and algorithms / Kernels and regression theory
Double descent and overfitting
Explain benign and harmful overfitting and double descent in overparameterized models.
53 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2018–2026) · an approach in Kernels and regression theory, keyword cluster “ridge · linear regression · regularization”
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.
19 papers in 2023–24 and 7 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 7.7% to 2.8%.
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
- 2018 · Overfitting or perfect fitting? Risk bounds for classification and regression rules that interpolate (NeurIPS 2018 · 140 citations)
- 2019 · Data-Dependence of Plateau Phenomenon in Learning with Neural Network --- Statistical Mechanical Analysis (NeurIPS 2019 · 11 citations)
- 2020 · An Investigation of Why Overparameterization Exacerbates Spurious Correlations (ICML 2020 · 106 citations)
- 2021 · Optimal Regularization can Mitigate Double Descent (ICLR 2021 · 46 citations)
- 2022 · Neural Collapse Under MSE Loss: Proximity to and Dynamics on the Central Path (ICLR 2022 · 19 citations)
- 2023 · A U-turn on Double Descent: Rethinking Parameter Counting in Statistical Learning (NeurIPS 2023 · 7 citations)
- 2024 · Model Collapse Demystified: The Case of Regression (NeurIPS 2024 · 6 citations)
- 2025 · Thumb on the Scale: Optimal Loss Weighting in Last Layer Retraining (NeurIPS 2025 · 1 citation)
- 2026 · Risk Phase Transitions in Spiked Regression: Alignment Driven Benign and Catastrophic Overfitting (ICLR 2026 · 0 citations)
Earliest papers here
- Overfitting or perfect fitting? Risk bounds for classification and regression rules that interpolate (NeurIPS 2018 · 140 citations)
- Data-Dependence of Plateau Phenomenon in Learning with Neural Network --- Statistical Mechanical Analysis (NeurIPS 2019 · 11 citations)
- An Investigation of Why Overparameterization Exacerbates Spurious Correlations (ICML 2020 · 106 citations)
What has been tried
45 papers here report 339 measured changes (a component added, removed or swapped, compared with a baseline): helped 55%, no clear effect 12%, hurt 24%, mixed 9%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Use independent design matrices across generations rather than reusing X_0 | Isotropic regression, T_0 < d · test error bias | (1-φ_0^{-n})||w_0||^2 | Hurt | Model Collapse Demystified: The Case of Regression (NeurIPS 2024 · 6 citations) |
| increase random-feature dimension | MNIST · test error | Mixed | High-Dimensional Analysis for Generalized Nonlinear Regression: From Asymptotics to Algorithm (AAAI 2024 · 3 citations) | |
| Increase the initialization scale | Modular addition test · time to sharp test-accuracy transition | alpha increases from 1 to 100 | Hurt | Dichotomy of Early and Late Phase Implicit Biases Can Provably Induce Grokking (ICLR 2024 · 2 citations) |
| Use a 128-dimensional latent space while retaining ρ̃ calculated from effective dimension 3 | CelebA Straight Hair · per-class error | No clear effect | Thumb on the Scale: Optimal Loss Weighting in Last Layer Retraining (NeurIPS 2025 · 1 citation) |
What reviewers will expect
Datasets and baselines named in the 5 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CIFAR10 · 2 of 5 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 26 papers; from OpenAlex and the papers' affiliations).
- Toyota Technological Institute at Chicago · 3 papers
- University of Oxford · 2 papers
- University of Chicago · 2 papers
- Google · 2 papers
- Weizmann Institute · 2 papers
- Weizmann Institute of Science · 2 papers
- University of British Columbia · 2 papers
Use it
Download BibTeX of 36 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
- High-dimensional asymptotic generalization (82 papers)
- High-dimensional linear regression theory (61 papers)
- Kernel ridge regression analysis (57 papers)
Elsewhere in the topic
- RKHS kernel methods (127 papers)
- Neural tangent kernel analysis (109 papers)
- Statistical perspectives on generalization (99 papers)
- Information-theoretic generalization bounds (94 papers)
- Kernel regression under shift (89 papers)
- Generalization theory of deep networks (67 papers)
- Empirical generalization measures (61 papers)
- PAC-Bayes bounds (39 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).