Research map / Graph learning / Graph algorithms and clustering
Embedded Feature Selection on Graph-Based Multi-View Clustering
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes EFSGMC, an anchor-graph multi-view clustering method that factorizes anchor graphs with NMF to get labels in one step, adds ℓ2,p-norm feature selection on anchors, and a tensor Schatten p-norm across view assignment matrices.
Problem
Anchor graph-based multi-view clustering typically needs post-processing (e.g., K-means) to get labels, does not exploit anchor graph information well, and underuses complementary inter-view information in large-scale data.
Approach
Factorize anchor graphs directly with non-negative matrix factorization to produce discrete cluster indicators in one step, perform anchor selection via row-sparsity (ℓ2,p-norm) on the latent matrix, add LPP-style label consistency, and couple views with a tensor Schatten p-norm over stacked cluster assignment matrices.
Key results
- On MSRC, EFSGMC reaches ACC/NMI/Purity of 1.000 versus 0.981/0.960/0.981 for MSC-BG.
- On Reuters, EFSGMC achieves ACC 0.618 and NMI 0.518, outperforming MSC-BG (0.640 ACC is MSC-BG's, so EFSGMC is not best in ACC there) and SFMC (0.602 ACC, 0.354 NMI).
- Several baselines (SwMC, SMSC, AMGL, MLAN, RMSC, CSMSC) run out of memory on Reuters.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use EFSGMC instead of MSC-BG | EFSGMC · MSRC | ACC: 0.981 -> 1.000 | Helped |
| Use EFSGMC instead of MSC-BG | EFSGMC · HW | ACC: 0.889 -> 0.994 | Helped |
| Use EFSGMC instead of MSC-BG | EFSGMC · Mnist | ACC: 0.938 -> 0.951 | Helped |
| Use EFSGMC instead of MSC-BG | EFSGMC · Cal101 | ACC: 0.667 -> 0.741 | Helped |
| Use EFSGMC instead of MSC-BG | EFSGMC · Reuters | ACC: 0.640 -> 0.618 | Hurt |
| Add tensor Schatten p-norm to EFSGMC (ablation) | EFSGMC · MSRC | clustering performance | Helped |
| Add tensor Schatten p-norm to EFSGMC (ablation) | EFSGMC · HW | clustering performance | Helped |
| Add tensor Schatten p-norm to EFSGMC (ablation) | EFSGMC · Mnist | clustering performance | Helped |
| Add tensor Schatten p-norm to EFSGMC (ablation) | EFSGMC · Cal101 | clustering performance | Helped |
| Vary the tensor Schatten p-norm parameter p from 0.1 to 1.0 (ablation) | EFSGMC · MSRC and Mnist | clustering performance | Mixed |
| Vary the anchor rate from 0.1 to 1.0 (ablation) | EFSGMC · MSRC and Mnist | clustering performance | Mixed |
| Use ℓ2,p-norm instead of ℓ2,1-norm for Hv (ablation) | EFSGMC · MSRC | feature-selection matrix sparsity | Helped |
Surprising findings
- Reported Reuters results show MSC-BG with higher ACC (0.640) than EFSGMC (0.618), while EFSGMC leads on NMI and Purity.
- The authors observe that using anchor rate 1.0 (all points as anchors) does not give the best performance, attributed to redundant data.
Limitations
- Many hyperparameters (γ, λ, β, p, anchor rate) tuned over grids, with optimal values chosen per dataset.
- Reported Reuters ACC is lower than MSC-BG's.
Experimental setup
Datasets: MSRC HW Mnist Cal101-20 Reuters
Benchmarks: MSRC HW Mnist Cal101-20 Reuters
Metrics: ACC NMI Purity
Baselines: Co-reg SwMC MVSC SMSC AMGL MLAN SFMC RMSC CSMSC MSC-BG FPMVS-CAG
Builds on: anchor graph clustering (SFMC) weighted tensor Schatten p-norm minimization locality preserving projection non-negative matrix factorization
Closest papers by meaning
- S2MVTC: a Simple yet Efficient Scalable Multi-View Tensor Clustering (CVPR 2024 · similarity 0.88)
- From Spectrum-free towards Baseline-view-free: Double-track Proximity Driven Multi-view Clustering (ICML 2025 · similarity 0.88)
- Fast Incomplete Multi-view Clustering by Flexible Anchor Learning (ICML 2025 · similarity 0.88)
- A Non-parametric Graph Clustering Framework for Multi-View Data (AAAI 2024 · similarity 0.87)
- Max-Mahalanobis Anchors Guidance for Multi-View Clustering (AAAI 2025 · similarity 0.87)
- Fast Incomplete Multi-view Clustering with Adaptive Similarity Completion and Reconstruction (AAAI 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.