Research map / Graph learning / Graph algorithms and clustering
A Non-parametric Graph Clustering Framework for Multi-View Data
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 NpGC, a hyper-parameter-free multi-view graph clustering framework that builds a consensus bipartite graph from view-related and view-unrelated anchors, with linear complexity, evaluated on six multi-view datasets against 17 baselines.
Problem
Multi-view graph clustering methods usually depend on at least one hyper-parameter, which complicates tuning, solving, and scalability to large datasets.
Approach
Replaces the hyper-parameter that balances reconstruction error and regularization with two anchor sets (view-related and view-unrelated) that are jointly learned and linked directly to a consensus bipartite graph built at anchor level, rather than merging per-view graphs.
Key results
- On Calte101view3, NpGC reaches 35.55% ACC, 28.12% NMI, 38.55% purity, outperforming the best baseline PFSC (35.41% ACC) and others, while having zero hyper-parameters.
- On STL10, NpGC achieves 87.46% ACC, behind UDBGL (83.29%) on NMI? No: NpGC gets 88.53% NMI vs UDBGL 88.99%, so UDBGL is second-best on NMI; NpGC is best on ACC and purity.
- NpGC handles datasets up to 126,054 samples (YTF50) with linear time and space complexity, while several baselines fail with N/A on the larger datasets.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add view-unrelated anchors to view-related anchors (ablation) | NpGC · Calte101view3 | ACC: 33.96±1.50 -> 35.55±2.25 | Helped |
| add view-related anchors to view-unrelated anchors (ablation) | NpGC · Calte101view3 | ACC: 32.95±3.37 -> 35.55±2.25 | Helped |
| add view-unrelated anchors to view-related anchors (ablation) | NpGC · CCV | ACC: 22.81±0.92 -> 23.39±0.96 | Helped |
| add view-related anchors to view-unrelated anchors (ablation) | NpGC · CCV | ACC: 23.01±0.75 -> 23.39±0.96 | Helped |
| add view-unrelated anchors to view-related anchors (ablation) | NpGC · STL10 | ACC: 83.99±7.66 -> 87.46±8.26 | Helped |
| add view-related anchors to view-unrelated anchors (ablation) | NpGC · STL10 | ACC: 86.13±3.91 -> 87.46±8.26 | Helped |
| add view-unrelated anchors to view-related anchors (ablation) | NpGC · VGGFace100 | ACC: 9.01±0.41 -> 9.28±0.32 | Helped |
| add view-related anchors to view-unrelated anchors (ablation) | NpGC · VGGFace100 | ACC: 8.44±0.31 -> 9.28±0.32 | Helped |
| add view-unrelated anchors to view-related anchors (ablation) | NpGC · YTF20 | ACC: 64.97±5.22 -> 65.76±5.13 | Helped |
| add view-related anchors to view-unrelated anchors (ablation) | NpGC · YTF20 | ACC: 64.32±5.33 -> 65.76±5.13 | Helped |
| add view-unrelated anchors to view-related anchors (ablation) | NpGC · YTF50 | ACC: 65.32±3.37 -> 66.13±3.01 | Helped |
| add view-related anchors to view-unrelated anchors (ablation) | NpGC · YTF50 | ACC: 64.99±3.19 -> 66.13±3.01 | Helped |
Surprising findings
- Several baselines (MVSC, AMGL, MLRSSC, MSCIAS, SFMC, PGSC, EEOMVC) cannot run on the larger datasets YTF20 and YTF50.
Limitations
- Second-best results on STL10 and YTF20/YTF50 relative to BMVC and UDBGL, which use view adjustment policies to weight views.
- Authors note that anchor weighting schemes are not yet designed and could further improve performance.
Experimental setup
Datasets: Caltech101-7 (Calte101view3) CCV STL10 VGGFace100 YTF20 YTF50
Benchmarks: Calte101view3 CCV STL10 VGGFace100 YTF20 YTF50
Metrics: ACC NMI Purity Fscore Precision running time
Baselines: MVSC AMGL MLRSSC MSCIAS BMVC FMR mPAC MCLES PMSC PFSC FMCNOF FPMVS
Builds on: anchor graph clustering spectral clustering
Closest papers by meaning
- Medusa: A Multi-Scale High-order Contrastive Dual-Diffusion Approach for Multi-View Clustering (CVPR 2025 · similarity 0.91)
- Anchors Bring Ease: An Embarrassingly Simple Approach to Partial Multi-View Clustering (AAAI 2019 · similarity 0.91)
- Incomplete and Unpaired Multi-View Graph Clustering with Cross-View Feature Fusion (AAAI 2025 · similarity 0.91)
- Attribute-Missing Multi-view Graph Clustering (CVPR 2025 · similarity 0.90)
- Learn from View Correlation: An Anchor Enhancement Strategy for Multi-view Clustering (CVPR 2024 · similarity 0.90)
- SURER: Structure-Adaptive Unified Graph Neural Network for Multi-View Clustering (AAAI 2024 · similarity 0.89)
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.