Research map / Probabilistic and scientific modeling / Variational and generative sampling
Regularizing VAE latent information
Controls information flow in latents via deterministic autoencoders, hierarchy and divergence regularizers.
89 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Variational and generative sampling, keyword cluster “vae · autoencoders · latent”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Quiet None of the tracked benchmarks has a report from 2025–26.
19 papers in 2023–24 and 13 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 4.6% to 2.6%.
No longer reported by papers here: CIFAR-10 (FID, last in 2022).
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
- 2017 · Adversarial Variational Bayes: Unifying Variational Autoencoders and Generative Adversarial Networks (ICML 2017 · 157 citations)
- 2018 · Wasserstein Auto-Encoders (ICLR 2018 · 517 citations)
- 2019 · Lagging Inference Networks and Posterior Collapse in Variational Autoencoders (ICLR 2019 · 207 citations)
- 2020 · NVAE: A Deep Hierarchical Variational Autoencoder (NeurIPS 2020 · 361 citations)
- 2021 · Simple and Effective VAE Training with Calibrated Decoders (ICML 2021 · 33 citations)
- 2022 · Fully Spiking Variational Autoencoder (AAAI 2022 · 44 citations)
- 2023 · Explicitly Minimizing the Blur Error of Variational Autoencoders (ICLR 2023 · 8 citations)
- 2024 · QVAE-Mole: The Quantum VAE with Spherical Latent Variable Learning for 3-D Molecule Generation (NeurIPS 2024 · 9 citations)
- 2025 · DR-VAE: Debiased and Representation-enhanced Variational Autoencoder for Collaborative Recommendation (AAAI 2025 · 8 citations)
- 2026 · Phase-Type Variational Autoencoders for Heavy-Tailed Data (ICML 2026 · 0 citations)
Earliest papers here
- Adversarial Variational Bayes: Unifying Variational Autoencoders and Generative Adversarial Networks (ICML 2017 · 157 citations)
- Learning Disentangled Representations with Semi-Supervised Deep Generative Models (NeurIPS 2017 · 140 citations)
- Denoising Criterion for Variational Auto-Encoding Framework (AAAI 2017 · 138 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, 2019–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 72.9 | Diagnosing and Enhancing VAE Models (ICLR 2019 · 89 citations) |
| 2021 | 12.19 | VAEBM: A Symbiosis between Variational Autoencoders and Energy-based Models (ICLR 2021 · 7 citations) |
What has been tried
75 papers here report 740 measured changes (a component added, removed or swapped, compared with a baseline): helped 66%, no clear effect 11%, hurt 18%, 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 |
|---|---|---|---|---|
| replace vMF latent distribution with a normal distribution | QM9 random generation · Valid | Hurt | QVAE-Mole: The Quantum VAE with Spherical Latent Variable Learning for 3-D Molecule Generation (NeurIPS 2024 · 9 citations) | |
| add the Flow-based Representation Enhancement module | Amazon Toys · Recall@20 | 0.0647 -> 0.1045 | Helped | DR-VAE: Debiased and Representation-enhanced Variational Autoencoder for Collaborative Recommendation (AAAI 2025 · 8 citations) |
| use straight-through gradients instead of exact gradients | van Hateren, linear encoder and decoder · percent drop in validation loss | 0.6 ± .5 -> 7.3 ± .1 | Hurt | Poisson Variational Autoencoder (NeurIPS 2024 · 5 citations) |
| Add the mixture-of-Wasserstein-barycenters aggregation to WB-VAE | MNIST-SVHN-TEXT · linear classification accuracy (Avg.) | 0.89 -> 0.97 | Helped | Multimodal Variational Autoencoder: A Barycentric View (AAAI 2025 · 4 citations) |
| make the encoder variance unlearnable | MNIST · AU | 69 % -> 100 % | Helped | Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 2024 · 2 citations) |
| replace norm scaling with the additive rotation formulation | ImageNet · r-IS | 106.3 -> 106.1 | No clear effect | Restructuring Vector Quantization with the Rotation Trick (ICLR 2025 · 1 citation) |
| Replace VDMv-8 with a small trainable encoder | CIFAR-10, fixed noise schedule · total BPD | 2.783± 0.004 -> 2.783± 0.004 | No clear effect | DiffEnc: Variational Diffusion with a Learned Encoder (ICLR 2024 · 1 citation) |
| Increase the number of HPA neighbours from 3 to 15 | Moving Ball trajectory reconstruction · RMSE | 17.400 ± 3.691 -> 1.709 ± 0.146 | Helped | Neighbour-Driven Gaussian Process Variational Autoencoders for Scalable Structured Latent Modelling (ICML 2025 · 0 citations) |
| Learn ELBO weights π and model expert dependence with ρ instead of equal weights and independent experts | MNIST-SVHN-Text · Overall performance across β values | Helped | Aggregation of Dependent Expert Distributions in Multimodal Variational Autoencoders (ICML 2025 · 0 citations) | |
| remove the HSIC penalty | Extended Yale B · identity accuracy | Hurt | StrWAEs to Invariant Representations (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
- PolyMNIST · 2 of 8 papers
- MNIST-SVHN-TEXT · 2 of 8 papers
Baselines
- MoPoE-VAE · 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 32 papers; from OpenAlex and the papers' affiliations).
- ETHZ - ETH Zurich · 2 papers
- Swiss Federal Institute of Technology · 2 papers
- University of Toronto · 2 papers
- National University of Singapore · 2 papers
- University of California, Berkeley · 2 papers
- Monash University · 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
- Structured latent VAEs (125 papers)
- Disentangled and discrete latent models (92 papers)
- Hierarchical VAEs and posterior collapse (59 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)
- Exact inference for probabilistic programs (103 papers)
- Latent variable marginalization and estimators (91 papers)
- Tractable probabilistic models (90 papers)
- Variational inference theory and gradients (89 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).