Research map / Probabilistic and scientific modeling / Molecules and proteins
Equivariant and disentangled latent representations
Learns group-structured, equivariant or disentangled latent spaces via symmetry priors.
90 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2018–2026) · an approach in Molecules and proteins, keyword cluster “latent · space · representations”
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.
27 papers in 2023–24 and 38 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 5.5% to 4.1%.
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 · Modeling Facial Geometry Using Compositional VAEs (CVPR 2018 · 124 citations)
- 2019 · Learning Disentangled Representations for Recommendation (NeurIPS 2019 · 99 citations)
- 2020 · Counterfactuals uncover the modular structure of deep generative models (ICLR 2020 · 28 citations)
- 2021 · The role of Disentanglement in Generalisation (ICLR 2021 · 22 citations)
- 2022 · Do Not Escape From the Manifold: Discovering the Local Coordinates on the Latent Space of GANs (ICLR 2022 · 8 citations)
- 2023 · Relative representations enable zero-shot latent space communication (ICLR 2023 · 7 citations)
- 2024 · A Generative Model of Symmetry Transformations (NeurIPS 2024 · 4 citations)
- 2025 · IsUMap: Manifold Learning and Data Visualization Leveraging Vietoris-Rips Filtrations (AAAI 2025 · 5 citations)
- 2026 · Geo2Vec: Shape- and Distance-Aware Neural Representation of Geospatial Entities (AAAI 2026 · 2 citations)
Earliest papers here
- Modeling Facial Geometry Using Compositional VAEs (CVPR 2018 · 124 citations)
- Learning Continuous Hierarchies in the Lorentz Model of Hyperbolic Geometry (ICML 2018 · 61 citations)
- MAGAN: Aligning Biological Manifolds (ICML 2018 · 33 citations)
What has been tried
61 papers here report 638 measured changes (a component added, removed or swapped, compared with a baseline): helped 64%, no clear effect 10%, hurt 21%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Use canonical t-conorm instead of algebraic sum | Hemisphere with non-uniform distribution · projection uniformity | Helped | IsUMap: Manifold Learning and Data Visualization Leveraging Vietoris-Rips Filtrations (AAAI 2025 · 5 citations) | |
| use the x-space SSL objective instead of the h-space objective | Fully rotated MNIST · x-mse | 0.3567 -> 0.2387 | Helped | A Generative Model of Symmetry Transformations (NeurIPS 2024 · 4 citations) |
| Add conformal isometry loss | Grid-cell response maps · Hexagonal pattern emergence | Helped | On Conformal Isometry of Grid Cells: Learning Distance-Preserving Position Embedding (ICLR 2025 · 3 citations) | |
| combine the individual TraceGrad mechanisms | Hamiltonian prediction · accuracy | Helped | TraceGrad: a Framework Learning Expressive SO(3)-equivariant Non-linear Representations for Electronic-Structure Hamiltonian Prediction (ICML 2025 · 2 citations) | |
| Remove equivariant tensor prediction | Dielectric test set · Fnorm | 3.76 -> 4.15 | Hurt | A Space Group Symmetry Informed Network for O(3) Equivariant Crystal Tensor Prediction (ICML 2024 · 2 citations) |
| remove static loss | PhysioNet · AUPRC | 0.274 ± 0.02 -> 0.473 ± 0.02 | Hurt | Sequential Disentanglement by Extracting Static Information From A Single Sequence Element (ICML 2024 · 1 citation) |
| Use approximate rather than exact fragment automorphism correction | Molecule generation (Fragment) · Top K div. | 0.151 -> 0.164 | Helped | Symmetry-Aware GFlowNets (ICML 2025 · 0 citations) |
| Use a 3-chart atlas instead of a 19-chart atlas | Heat equation PDE · Symmetry recovery | Hurt | AtlasD: Automatic Local Symmetry Discovery (ICML 2025 · 0 citations) | |
| Replace the SO(3)-equivariant Gram-Schmidt local frame with quaternion-based SPFrame | JARVIS-DFT formation energy · MAE (eV/atom) | 0.0285 -> 0.0261 | Helped | Local-Global Associative Frames for Symmetry-Preserving Crystal Structure Modeling (NeurIPS 2025 · 0 citations) |
| Initialize γ to 1 rather than 0 | N-Body · test MSE | 5.39 ± 0.33 -> 4.97 ± 0.01 | Helped | Learning (Approximately) Equivariant Networks via Constrained Optimization (NeurIPS 2025 · 0 citations) |
| Replace autoregressive lattice sampling with joint lattice and coordinate diffusion | MP-20 · Wasserstein distance for number of elements | 0.2092 -> 0.1292 | Helped | Space Group Equivariant Crystal Diffusion (NeurIPS 2025 · 0 citations) |
| use a symmetric loss | O3 and GL3 synthetic data · Acc(G, 5) | O3, negatives both: 89.75 -> 99.83; GL3, negatives both: 99.98 -> 99.93 | Helped | Equivariance by Contrast: Identifiable Equivariant Embeddings from Unlabeled Finite Group Actions (NeurIPS 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 22 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- MNIST · 3 of 22 papers
- QM9 · 3 of 22 papers
- CUB · 2 of 22 papers
Baselines
- EGNN · 2 of 22 papers
- SchNet · 2 of 22 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 65 papers; from OpenAlex and the papers' affiliations).
- Northeastern University · 5 papers
- University of Amsterdam · 4 papers
- Stanford University · 3 papers
- Texas A&M University - College Station · 3 papers
- Amazon · 2 papers
- State University of New York · 2 papers
- Shanghai Jiao Tong University · 2 papers
- University of California, San Diego · 2 papers
- IBM · 2 papers
- University of Trento · 2 papers
Use it
Download BibTeX of 39 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
- Neuro-symbolic probabilistic layers (120 papers)
- Sparse graphical and tensor models (97 papers)
- Probabilistic embeddings and relational inference (93 papers)
- Generative latent models for single-cell data (41 papers)
Elsewhere in the topic
- Generative protein structure design (142 papers)
- Prediction-powered and conformal inference (129 papers)
- Statistical analysis of ML practice (122 papers)
- Structure-based 3D molecule generation (107 papers)
- Flow matching generative modeling (97 papers)
- Machine learning for physical simulation (96 papers)
- Multimodal trajectory prediction (95 papers)
- Counterfactual and statistical explanations (81 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).