Research map / Learning theory and algorithms / Clustering
TPCH: Tensor-interacted Projection and Cooperative Hashing for 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 TPCH, a tensor-interacted projection and hashing method for large-scale multi-view clustering that stacks projection and hash matrices into tensors with an enhanced tensor nuclear norm, achieving better clustering and much lower CPU time than GCAE on five datasets.
Problem
Large-scale multi-view clustering with binary hash codes suffers from independent projection matrices that lack inter-view communication and higher-order collaboration, degrading hash compactness, clustering quality, and noise robustness.
Approach
Stacks per-view projection matrices and hash matrices into tensors, applies an enhanced tensor nuclear norm (ETNN) that additionally low-rank-regularizes the t-SVD core tensor, and learns hash codes in Hamming space with discrete clustering via Hamming distance.
Key results
- On Cifar-10, TPCH reports ACC 0.70 and NMI 0.67 versus best baseline ACC of about 0.29 and NMI 0.13 in the table.
- CPU time vs GCAE: TPCH is 7.1x to 10.2x faster across the five datasets (e.g., Caltech101 114.11 s vs 1158.71 s).
- TPCH achieves best ACC on 4 of 5 datasets in Table 1 except 100 leaves, where GCAE (0.89) and AC-MVBC/mPAC edge out TPCH (0.86).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use TPCH instead of the strongest listed baseline | Multi-view clustering · SUNRGBD | ACC: 0.24 -> 0.22 | Hurt |
| Use TPCH instead of the strongest listed baseline | Multi-view clustering · Cifar-10 | ACC: 0.29 -> 0.70 | Helped |
| Use TPCH instead of the strongest listed baseline | Multi-view clustering · Caltech101 | ACC: 0.32 -> 0.54 | Helped |
| Use TPCH instead of the strongest listed baseline | Multi-view clustering · Caltech256 | ACC: 0.11 -> 0.13 | Helped |
| Use TPCH instead of the strongest listed baseline | Multi-view clustering · 100 leaves | ACC: 0.89 -> 0.86 | Hurt |
| Use TPCH instead of the strongest listed baseline | Multi-view clustering · SUNRGBD | NMI: 0.24 -> 0.28 | Helped |
| Use TPCH instead of the strongest listed baseline | Multi-view clustering · Cifar-10 | NMI: 0.13 -> 0.67 | Helped |
| Use TPCH instead of the strongest listed baseline | Multi-view clustering · Caltech101 | NMI: 0.51 -> 0.80 | Helped |
| Use TPCH instead of the strongest listed baseline | Multi-view clustering · Caltech256 | NMI: 0.32 -> 0.37 | Helped |
| Use TPCH instead of the strongest listed baseline | Multi-view clustering · 100 leaves | NMI: 0.94 -> 0.97 | Helped |
| Use TPCH instead of the strongest listed baseline | Multi-view clustering · SUNRGBD | Purity: 0.35 -> 0.37 | Helped |
| Use TPCH instead of the strongest listed baseline | Multi-view clustering · Cifar-10 | Purity: 0.27 -> 0.73 | Helped |
Surprising findings
- On 100 leaves, TPCH (ACC 0.86) is not the best; GCAE (0.89) and mPAC-related baselines are higher on that dataset in the reported table.
Limitations
- Single hyperparameter alpha, but results depend on anchor count m and kernel width
- Comparisons based on MATLAB implementations of baselines using their public code; reported runtime only versus GCAE
- Some reported numbers (e.g., Cifar-10 gains) are very large relative to baselines, raising questions about comparability
Experimental setup
Datasets: SUNRGBD Cifar-10 Caltech101 Caltech256 100 leaves Synthetic_4clu Synthetic_3d
Benchmarks: SUNRGBD Cifar-10 Caltech101 Caltech256 100 leaves
Metrics: ACC NMI Purity F-score ARI CPU time
Baselines: SC Co-regularize AMGL Mul-NMF MLAN mPAC GMC SH DSH SP ITQ SGH
Builds on: AC-MVBC binary multi-view clustering tensor nuclear norm
Closest papers by meaning
- S2MVTC: a Simple yet Efficient Scalable Multi-View Tensor Clustering (CVPR 2024 · similarity 0.89)
- Scalable Federated One-Step Multi-View Clustering with Tensorized Regularization (AAAI 2025 · similarity 0.87)
- Performance Gaps in Multi-view Clustering under the Nested Matrix-Tensor Model (ICLR 2024 · similarity 0.87)
- Large-scale Multi-view Tensor Clustering with Implicit Linear Kernels (CVPR 2025 · similarity 0.87)
- EDISON: Enhanced Dictionary-Induced Tensorized Incomplete Multi-View Clustering with Gaussian Error Rank Minimization (ICML 2024 · similarity 0.86)
- From Dictionary to Tensor: A Scalable Multi-View Subspace Clustering Framework with Triple Information Enhancement (NeurIPS 2024 · similarity 0.86)
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.