Research map / Probabilistic and scientific modeling / Molecules and proteins
Sparse graphical and tensor models
Learns sparse graphs and tensor decompositions for high-order interaction data.
97 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · 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.
16 papers in 2023–24 and 38 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 3.3% 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
- 2016 · Discriminative Embeddings of Latent Variable Models for Structured Data (ICML 2016 · 499 citations)
- 2017 · Learning to Discover Sparse Graphical Models (ICML 2017 · 11 citations)
- 2018 · The Sparse Manifold Transform (NeurIPS 2018 · 17 citations)
- 2019 · Learning Phenotypes and Dynamic Patient Representations via RNN Regularized Collective Non-Negative Tensor Factorization (AAAI 2019 · 18 citations)
- 2020 · Efficient Derivative Computation for Cumulative B-Splines on Lie Groups (CVPR 2020 · 95 citations)
- 2021 · STELAR: Spatio-temporal Tensor Factorization with Latent Epidemiological Regularization (AAAI 2021 · 15 citations)
- 2022 · HLRTF: Hierarchical Low-Rank Tensor Factorization for Inverse Problems in Multi-Dimensional Imaging (CVPR 2022 · 61 citations)
- 2023 · LORE: Logical Location Regression Network for Table Structure Recognition (AAAI 2023 · 36 citations)
- 2024 · Approximating mutual information of high-dimensional variables using learned representations (NeurIPS 2024 · 11 citations)
- 2025 · Skeleton-based Action Recognition with Non-linear Dependency Modeling and Hilbert-Schmidt Independence Criterion (AAAI 2025 · 7 citations)
- 2026 · SMART: Scalable Mesh‑free Aerodynamic Simulations from Raw Geometries using a Transformer‑based Surrogate Model (ICML 2026 · 0 citations)
Earliest papers here
- Discriminative Embeddings of Latent Variable Models for Structured Data (ICML 2016 · 499 citations)
- Learning Sparse Gaussian Graphical Models with Overlapping Blocks (NeurIPS 2016 · 17 citations)
- Hierarchical Compound Poisson Factorization (ICML 2016 · 13 citations)
What has been tried
77 papers here report 701 measured changes (a component added, removed or swapped, compared with a baseline): helped 69%, no clear effect 9%, hurt 16%, 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 cross-predictive regularization with MINE regularization | Synthetic Gaussian benchmarks · absolute and relative accuracy | Hurt | Approximating mutual information of high-dimensional variables using learned representations (NeurIPS 2024 · 11 citations) | |
| Vary τ from 0.1 to 0.2 | FAUST Remeshed · Accuracy (r = 0) | 62.61 -> 62.58 | No clear effect | Finsler-Laplace-Beltrami Operators with Application to Shape Analysis (CVPR 2024 · 9 citations) |
| Add joint and bone inputs rather than using joint input only | NTU RGB+D 120 X-Sub · accuracy | 86.0 -> 89.3 | Helped | Skeleton-based Action Recognition with Non-linear Dependency Modeling and Hilbert-Schmidt Independence Criterion (AAAI 2025 · 7 citations) |
| Replace Ct-Echo-Decay with Ct-Echo-Leaky | CWRU datasets, 50% missing · accuracy (%) | 75.22 ± 2.10 -> 83.06 ± 0.88; 75.78 ± 1.66 -> 82.83 ± 1.06; 80.72 ± 2.36 -> 87.67 ± 1.20; 86.67 ± 1.22 -> 92.67 ± 0.81; | Helped | Efficient Anomaly Detection of Irregular Sequences in Ct-Echo Model Space (AAAI 2025 · 2 citations) |
| Add scanning integration to convolution integration | IXI radial ×4 · PSNR (dB) | 34.68 -> 48.16 | Helped | LMO: Linear Mamba Operator for MRI Reconstruction (CVPR 2025 · 2 citations) |
| Increase projector depth beyond three hidden layers | MNIST · 10-NN accuracy | No clear effect | Navigating the Effect of Parametrization for Dimensionality Reduction (NeurIPS 2024 · 2 citations) | |
| Replace compact pseudo-label reconstruction in Eq. (5) with relaxed homogeneity reconstruction in Eq. (4) | Patent · NMI | 34.21 -> 36.30 | Helped | Community-Aware Variational Autoencoder for Continuous Dynamic Networks (AAAI 2025 · 1 citation) |
| add reward-guided DPO post-training | Art-XL2.0 · IoU | 82.183% -> 82.806% | Helped | Auto-Connect: Connectivity-Preserving RigFormer with Direct Preference Optimization (NeurIPS 2025 · 1 citation) |
| replace TwoNN with MLE for intrinsic-dimension estimation | Complete nonlinear correlation · I_dCor | 0.67±0.02 -> 0.38±0.01 | Hurt | Intrinsic Dimension Correlation: uncovering nonlinear connections in multimodal representations (ICLR 2025 · 1 citation) |
| use k=16 eigen-functions instead of the default | GEOM-QM9 · Recall AMR (mean) | 0.124 -> 0.139 | Hurt | Manifold Diffusion Fields (ICLR 2024 · 1 citation) |
| Set Eigen-SAM alignment strength to α=0.2 instead of α=0 | CIFAR-100 test · accuracy | 0.9% improvement | Helped | Explicit Eigenvalue Regularization Improves Sharpness-Aware Minimization (NeurIPS 2024 · 1 citation) |
| use a disconnected, partitioned fundamental region instead of the connected region | Synthetic SO(2) field F2 · RMSE | 0.13 [0.05] -> 0.49 | Hurt | Integration-free Kernels for Equivariant Gaussian Process Modelling (ICML 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 25 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 · 2 of 25 papers
- CIFAR10 · 2 of 25 papers
- ImageNet · 2 of 25 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 54 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 3 papers
- University of California, Los Angeles · 3 papers
- Zhejiang University · 2 papers
- University of Science and Technology of China · 2 papers
- Yale University · 2 papers
- University of Utah · 2 papers
- Harvard University · 2 papers
Use it
Download BibTeX of 38 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)
- Probabilistic embeddings and relational inference (93 papers)
- Equivariant and disentangled latent representations (90 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).