Research map / Learning theory and algorithms / Clustering
Subspace clustering
Cluster data lying near union of subspaces using self-expression and multi-view methods.
102 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Clustering, keyword cluster “correlation clustering · hierarchical · spectral clustering”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 2 of 4 tracked benchmarks.
17 papers in 2023–24 and 15 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 12.1% to 7.3%.
Open questions the data points to
- AwA (ACC): best went from 9.65 (up to 2024) to 10.65 in 2025 — Large-scale Multi-view Tensor Clustering with Implicit Linear Kernels (CVPR 2025 · 6 citations)
- AwA (NMI): best went from 11.52 (up to 2024) to 12.31 in 2025 — Large-scale Multi-view Tensor Clustering with Implicit Linear Kernels (CVPR 2025 · 6 citations)
- Papers report BMVC on AwA (ACC) anywhere from 8.66 to 10.45: comparisons against it depend on whose number you take.
What looks settled
- Mfeat (ACC (%)): best up to 2024 96.95, best in 2025–26 94.7 from 2 papers, at most 0.2% better, if at all.
- Mfeat (NMI (%)): best up to 2024 93.42, best in 2025–26 90.7 from 2 papers, at most 0.2% better, if at all.
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 · Scalable Sparse Subspace Clustering by Orthogonal Matching Pursuit (CVPR 2016 · 394 citations)
- 2017 · Superpixels and Polygons Using Simple Non-Iterative Clustering (CVPR 2017 · 496 citations)
- 2018 · SpectralNet: Spectral Clustering using Deep Neural Networks (ICLR 2018 · 115 citations)
- 2019 · Efficient and Effective Incomplete Multi-View Clustering (AAAI 2019 · 89 citations)
- 2020 · Tensor-SVD Based Graph Learning for Multi-View Subspace Clustering (AAAI 2020 · 221 citations)
- 2021 · Hierarchical Multiple Kernel Clustering (AAAI 2021 · 50 citations)
- 2022 · Eigenlanes: Data-Driven Lane Descriptors for Structurally Diverse Lanes (CVPR 2022 · 60 citations)
- 2023 · Auto-Weighted Multi-View Clustering for Large-Scale Data (AAAI 2023 · 98 citations)
- 2024 · Correspondence-Free Non-Rigid Point Set Registration Using Unsupervised Clustering Analysis (CVPR 2024 · 36 citations)
- 2025 · Sharper Error Bounds in Late Fusion Multi-view Clustering with Eigenvalue Proportion Optimization (AAAI 2025 · 6 citations)
- 2026 · Deep Clustering Based on Sparse Kolmogorov-Arnold Network and Spectral Constraint (AAAI 2026 · 1 citation)
Earliest papers here
- Scalable Sparse Subspace Clustering by Orthogonal Matching Pursuit (CVPR 2016 · 394 citations)
- Oracle Based Active Set Algorithm for Scalable Elastic Net Subspace Clustering (CVPR 2016 · 247 citations)
- Manifold SLIC: A Fast Method to Compute Content-Sensitive Superpixels (CVPR 2016 · 164 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.
AwA · ACC (higher is better)
5 papers, 2022–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 9.65 | Stability and Generalization of Kernel Clustering: from Single Kernel to Multiple Kernel (NeurIPS 2022 · 0 citations) |
| 2025 | 10.65 | Large-scale Multi-view Tensor Clustering with Implicit Linear Kernels (CVPR 2025 · 6 citations) |
AwA · NMI (higher is better)
4 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 10.89 | Consistency of Multiple Kernel Clustering (ICML 2023) |
| 2024 | 11.52 | Scalable Multiple Kernel Clustering: Learning Clustering Structure from Expectation (ICML 2024) |
| 2025 | 12.31 | Large-scale Multi-view Tensor Clustering with Implicit Linear Kernels (CVPR 2025 · 6 citations) |
Mfeat · ACC (%) (higher is better)
4 papers, 2022–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 96.95 | Robust Graph-Based Multi-View Clustering (AAAI 2022 · 33 citations) |
Mfeat · NMI (%) (higher is better)
4 papers, 2022–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 93.42 | Robust Graph-Based Multi-View Clustering (AAAI 2022 · 33 citations) |
What has been tried
94 papers here report 846 measured changes (a component added, removed or swapped, compared with a baseline): helped 74%, no clear effect 9%, hurt 13%, 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 |
|---|---|---|---|---|
| Replace the Gaussian kernel with the Laplacian kernel | Bunny registration under varying noise · RMSE | Helped | Correspondence-Free Non-Rigid Point Set Registration Using Unsupervised Clustering Analysis (CVPR 2024 · 36 citations) | |
| vary the low-rank constraint weight λ | HW and CCV · ACC | Mixed | Low-Rank Kernel Tensor Learning for Incomplete Multi-View Clustering (AAAI 2024 · 35 citations) | |
| Use diverse view-shared anchors with multi-dimension and multi-size instead of single-dimension single-size anchors | Caltech101-7, missing rate 0.1 · ACC | 51.22 -> 62.73 | Helped | DVSAI: Diverse View-Shared Anchors Based Incomplete Multi-View Clustering (AAAI 2024 · 24 citations) |
| Use leverage score sampling instead of uniform sampling | CUB · ACC | 78.83 -> 81.33 | Helped | Incremental Nyström-based Multiple Kernel Clustering (AAAI 2025 · 5 citations) |
| Add the missing kernel elements recovery strategy | ORL and Caltech101-7 · clustering performance | Helped | KOALA: Kernel Coupling and Element Imputation Induced Multi-View Clustering (AAAI 2025 · 5 citations) | |
| add HTR | NGs · ACC | 0.379±0.000 -> 1.000±0.000 | Helped | From Dictionary to Tensor: A Scalable Multi-View Subspace Clustering Framework with Triple Information Enhancement (NeurIPS 2024 · 2 citations) |
| Add SCE to DMJC | Caltech101-7 · ACC | 0.469 -> 0.583 | Helped | Evaluate then Cooperate: Shapley-based View Cooperation Enhancement for Multi-view Clustering (NeurIPS 2024 · 2 citations) |
| Vary λ across the tested range | Eight multi-view datasets · Clustering performance (ACC, NMI, F1-score) | No clear effect | Robust Consensus Anchor Learning for Efficient Multi-view Subspace Clustering (ICML 2025 · 0 citations) | |
| Replace linearly distance-proportional weights with exponential distance weights | CIFAR-10 · ACC | 96.07 -> 94.34 | Hurt | Geometric Analysis of Nonlinear Manifold Clustering (NeurIPS 2024 · 0 citations) |
| Use nucleus sampling with t=0.1 instead of t=1 | Erdős-Rényi graphs, n=3–5 · approximation ratio | 0.97 ± 0.04 -> 1.00 ± 0.00; 0.92 ± 0.07 -> 1.00 ± 0.00; 0.92 ± 0.07 -> 1.00 ± 0.01 | Helped | A Sub-Problem Quantum Alternating Operator Ansatz for Correlation Clustering (ICML 2025) |
| Add EDR | NGs · ACC | 0.632 -> 1.000 | Helped | EDISON: Enhanced Dictionary-Induced Tensorized Incomplete Multi-View Clustering with Gaussian Error Rank Minimization (ICML 2024) |
| Increase the number of sampled anchors | Handwritten and ALOI-100 · Subspace distance | Helped | Scalable Multiple Kernel Clustering: Learning Clustering Structure from Expectation (ICML 2024) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| BMVC | AwA | ACC | 8.66 – 10.45 | 2 papers, e.g. NeurIPS 2022 and CVPR 2025 |
| t-SVD-MSC | Scene-15 | ACC | 0.812 – 0.892 | 2 papers, e.g. AAAI 2020 and AAAI 2020 |
| t-SVD-MSC | Caltech-101 | ACC | 0.556 – 0.607 | 2 papers, e.g. AAAI 2020 and AAAI 2020 |
| ETLMSC | HW | ACC | 0.938 – 0.998 | 2 papers, e.g. AAAI 2023 and NeurIPS 2024 |
What reviewers will expect
Datasets and baselines named in the 9 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- ORL · 3 of 9 papers
- ALOI · 2 of 9 papers
- Caltech101-20 · 2 of 9 papers
- Caltech101-all · 2 of 9 papers
- Mfeat · 2 of 9 papers
- AwA · 2 of 9 papers
Baselines
- LMVSC · 2 of 9 papers
- OMSC · 2 of 9 papers
- BMVC · 2 of 9 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).
- National University of Defense Technology · 7 papers
- Beijing Jiaotong University · 4 papers
- Nanjing University of Aeronautics and Astronautics · 3 papers
- Xidian University · 2 papers
- Zhejiang Lab · 2 papers
- Intelligent Game and Decision Lab · 2 papers
- Beijing Open University · 2 papers
- University of Southern California · 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
- Clustering approximation algorithms (150 papers)
- Correlation clustering (102 papers)
- Fairness-constrained clustering (19 papers)
- Incomplete multi-view clustering (125 papers)
- Spectral clustering algorithms (43 papers)
Elsewhere in the topic
- Coresets and dynamic clustering (77 papers)
- k-means++ seeding analysis (76 papers)
- Fair clustering approximation (31 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).