Research map / Probabilistic and scientific modeling / Variational and generative sampling
Normalizing flow training and design
Builds normalizing flows with stochastic interpolants, dense connections and symmetry equivariance.
88 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Variational and generative sampling, keyword cluster “flows · normalizing · invertible”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Quiet None of the tracked benchmarks has a report from 2025–26.
27 papers in 2023–24 and 10 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 6.5% to 2.0%.
No longer reported by papers here: CIFAR-10 (FID, last in 2023).
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 · Improved Variational Inference with Inverse Autoregressive Flow (NeurIPS 2016 · 762 citations)
- 2018 · Continuous-Time Flows for Efficient Inference and Density Estimation (ICML 2018 · 29 citations)
- 2019 · Discrete Flows: Invertible Generative Models of Discrete Data (NeurIPS 2019 · 66 citations)
- 2020 · Normalizing Flows on Tori and Spheres (ICML 2020 · 71 citations)
- 2021 · OT-Flow: Fast and Accurate Continuous Normalizing Flows via Optimal Transport (AAAI 2021 · 57 citations)
- 2022 · Path-Gradient Estimators for Continuous Normalizing Flows (ICML 2022 · 6 citations)
- 2023 · Building Normalizing Flows with Stochastic Interpolants (ICLR 2023 · 23 citations)
- 2024 · Robustly Train Normalizing Flows via KL Divergence Regularization (AAAI 2024 · 3 citations)
- 2025 · Path Gradients after Flow Matching (NeurIPS 2025 · 1 citation)
- 2026 · Bidirectional Normalizing Flow: From Data to Noise and Back (CVPR 2026 · 0 citations)
Earliest papers here
- Improved Variational Inference with Inverse Autoregressive Flow (NeurIPS 2016 · 762 citations)
- Continuous-Time Flows for Efficient Inference and Density Estimation (ICML 2018 · 29 citations)
- Neural Autoregressive Flows (ICML 2018 · 0 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
CIFAR-10 · FID (lower is better)
4 papers, 2021–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 13.43 | Diffusion Normalizing Flow (NeurIPS 2021 · 0 citations) |
| 2023 | 10.27 | Building Normalizing Flows with Stochastic Interpolants (ICLR 2023 · 23 citations) |
What has been tried
81 papers here report 864 measured changes (a component added, removed or swapped, compared with a baseline): helped 65%, no clear effect 9%, hurt 21%, 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 |
|---|---|---|---|---|
| Vary the KLJ coefficient α | Rot-RGB CIFAR10, TinyImageNet, QuickDraw, and Noisy CIFAR10 · OoD detection score | Helped | Robustly Train Normalizing Flows via KL Divergence Regularization (AAAI 2024 · 3 citations) | |
| Fine-tune the OT FM-pretrained CNF with Path Gradients | LJ13 · ESS_q (%) | 54.36 ± 5.43 -> 82.97 ± 0.40 | Helped | Path Gradients after Flow Matching (NeurIPS 2025 · 1 citation) |
| Use the off-manifold encoder Jacobian rather than the on-manifold encoder Jacobian | Tabular data · FID-like metric | Helped | Lifting Architectural Constraints of Injective Flows (ICLR 2024 · 1 citation) | |
| Increase choice-set size from k=2 to k=10 | Onemoon2D (n=100) · Wasserstein distance | 0.70 -> 0.11 | Helped | Preferential Normalizing Flows (NeurIPS 2024 · 1 citation) |
| Use TTF rather than TTFfix, allowing tail parameters to be learned | Insurance, Fama 5, S&P 500, CLIMDEX · mean test negative log likelihood | 1.38 (0.01) -> 1.37 (0.02); 4.63 (0.01) -> 4.61 (0.01); 314.84 (0.46) -> 317.56 (0.56); -2090.91 (11.90) -> -2214.28 (13 | Mixed | Flexible Tails for Normalizing Flows (ICML 2025 · 0 citations) |
| Replace the endpoint negative-log-density term with its first-order Taylor approximation | Gaussian Mixture Model, Funnel, and ExpGauss experiments · sampling performance | Helped | Annealing Flow Generative Models Towards Sampling High-Dimensional and Multi-Modal Distributions (ICML 2025 · 0 citations) | |
| Add importance-based rejection steps to neural JKO | Mustache · energy distance | 1.8e-2 -> 2.9e-3 | Helped | Importance Corrected Neural JKO Sampling (ICML 2025 · 0 citations) |
| Remove auxiliary points | Miniboone, 500 examples · log-likelihood | -27.75 -> -41.83 | Hurt | Kernelised Normalising Flows (ICLR 2024 · 0 citations) |
| Remove the highest 5 loss values from each training batch | Alanine dipeptide · Training stability | Helped | Conditional Normalizing Flows for Active Learning of Coarse-Grained Molecular Representations (ICML 2024 · 0 citations) | |
| Use RK4 with NFE=100 instead of Dopri5 | CIFAR-10 32×32 · FID | 4.76 vs 5.37 | Hurt | Reflected Flow Matching (ICML 2024 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 7 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Baselines
- CRAFT · 2 of 7 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 37 papers; from OpenAlex and the papers' affiliations).
- Freie Universität Berlin · 4 papers
- Mohamed bin Zayed University of Artificial Intelligence · 2 papers
- Peking University · 2 papers
- Hasso Plattner Institute · 2 papers
- TU Kaiserslautern · 2 papers
- Ruprecht-Karls-Universität Heidelberg · 2 papers
- Heidelberg University · 2 papers
- Tsinghua University · 2 papers
- Georgia Institute of Technology · 2 papers
Use it
Download BibTeX of 40 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
- Expressive invertible architectures (76 papers)
- Flow-based density estimation (29 papers)
Elsewhere in the topic
- Likelihood-free and graphical model estimation (161 papers)
- Flow matching and flow sampling (144 papers)
- Posterior sampling with diffusion guidance (137 papers)
- Score-based generative modeling analysis (128 papers)
- Structured latent VAEs (125 papers)
- Exact inference for probabilistic programs (103 papers)
- Disentangled and discrete latent models (92 papers)
- Latent variable marginalization and estimators (91 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).