Research map / Learning theory and algorithms / Prediction and decision losses
Loss landscape analysis
Characterize local minima and critical points in neural network loss surfaces.
61 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Prediction and decision losses, keyword cluster “losses · loss functions · surrogate”
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.
11 papers in 2023–24 and 11 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 2.1% to 1.9%.
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
- 2016 · Learning Deep Embeddings with Histogram Loss (NeurIPS 2016 · 257 citations)
- 2017 · Optimal Binary Autoencoding with Pairwise Correlations (ICLR 2017 · 0 citations)
- 2018 · Deep Linear Networks with Arbitrary Loss: All Local Minima Are Global (ICML 2018 · 80 citations)
- 2019 · Neural Persistence: A Complexity Measure for Deep Neural Networks Using Algebraic Topology (ICLR 2019 · 61 citations)
- 2020 · Learning to Autofocus (CVPR 2020 · 72 citations)
- 2021 · Interpreting and Boosting Dropout from a Game-Theoretic View (ICLR 2021 · 7 citations)
- 2022 · Critical Points in Quantum Generative Models (ICLR 2022 · 11 citations)
- 2023 · Deconstructing Data Reconstruction: Multiclass, Weight Decay and General Losses (NeurIPS 2023 · 3 citations)
- 2024 · Zipfian Whitening (NeurIPS 2024 · 3 citations)
- 2025 · HotSpot: Signed Distance Function Optimization with an Asymptotically Sufficient Condition (CVPR 2025 · 10 citations)
- 2026 · Deep Flow Networks (ICML 2026 · 0 citations)
Earliest papers here
- Learning Deep Embeddings with Histogram Loss (NeurIPS 2016 · 257 citations)
- Optimal Binary Autoencoding with Pairwise Correlations (ICLR 2017 · 0 citations)
- How regularization affects the critical points in linear networks (NeurIPS 2017 · 0 citations)
What has been tried
52 papers here report 469 measured changes (a component added, removed or swapped, compared with a baseline): helped 61%, no clear effect 11%, hurt 23%, 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 the heat loss to boundary and eikonal losses | 2D dataset · IoU | 0.8192 -> 0.9851 | Helped | HotSpot: Signed Distance Function Optimization with an Asymptotically Sufficient Condition (CVPR 2025 · 10 citations) |
| Remove whitening, retaining Zipfian centering only | STS-B · Spearman correlation ×100 | 66.92 -> 52.25 | Hurt | Zipfian Whitening (NeurIPS 2024 · 3 citations) |
| Use wavelet order 3 | Spleen test set · Dice score | 0.921 (6.91 · 10−2) -> 0.911 (7.38 · 10−2) | Hurt | Task-Driven Wavelets using Constrained Empirical Risk Minimization (CVPR 2024 · 1 citation) |
| Increase embedding dimensions | CUB · MAP@R | Helped | Mean Field Theory in Deep Metric Learning (ICLR 2024 · 1 citation) | |
| Add sparse training examples using f_inv_sqrt instead of default sampling | Modular addition N=16, q=257 · τ=0.5% accuracy | 1.2% -> 99.8% | Helped | Making Hard Problems Easier with Custom Data Distributions and Loss Regularization: A Case Study in Modular Arithmetic (ICML 2025 · 0 citations) |
| Add importance-weighting drift (ξ=1) rather than removing it | Waterbirds · worst-group accuracy | 81.7_{±{2.6}} -> 84.1_{±{1.1}} | Helped | Elliptic Loss Regularization (ICLR 2025 · 0 citations) |
| Replace no normalization with batch normalization or layer normalization | MNIST rotations and CIFAR10 noise · reversion-to-OCS trend | No clear effect | Deep Neural Networks Tend To Extrapolate Predictably (ICLR 2024 · 0 citations) | |
| Remove the eikonal loss | Unit sphere · flatness error | Hurt | Polyhedral Complex Derivation from Piecewise Trilinear Networks (NeurIPS 2024 · 0 citations) | |
| allow dual couplings ˜β and ˜κ to vary | 2D Ising model with target κ = 0 · recovery of expected Kramers-Wannier duals | Helped | A Machine Learning Approach to Duality in Statistical Physics (ICML 2025) | |
| Vary distortion parameter c across 0.95, 0.65, and 0.35 | CIFAR-10 and CIFAR-100 · accuracy | Mixed | Monotone, Bi-Lipschitz, and Polyak-Łojasiewicz Networks (ICML 2024) |
What reviewers will expect
Datasets and baselines named in the 8 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 8 papers
- CIFAR-100 · 2 of 8 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 22 papers; from OpenAlex and the papers' affiliations).
- University of Amsterdam · 2 papers
- Google · 2 papers
- Massachusetts Institute of Technology · 2 papers
- Whiting School of Engineering · 2 papers
- Dataheroes · 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
- Consistency of surrogate losses (72 papers)
- Losses for decision-making (68 papers)
- Ranking-based loss optimization (50 papers)
Elsewhere in the topic
- Adaptive testing and experiments (201 papers)
- Active learning and subset selection (199 papers)
- Distributionally robust optimization (184 papers)
- Data attribution and recourse (160 papers)
- Theory of neural network sensitivity (143 papers)
- Ranking models from comparisons (127 papers)
- Determinantal point process inference (122 papers)
- Conformal prediction sets (116 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).