Research map / Learning theory and algorithms / Clustering
Low-Rank Kernel Tensor Learning for Incomplete 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 LRKT-IMVC, which stacks incomplete kernels into a kernel tensor and imputes missing kernel entries with a low-rank tensor nuclear norm constraint, jointly with kernel k-means clustering via ADMM-style alternating optimization.
Problem
Incomplete multi-view clustering in the kernel setting, where existing incomplete multiple kernel clustering methods impute each incomplete kernel independently and ignore inter-view consistency and global kernel structure.
Approach
Stacks the per-view kernel matrices into a kernel tensor and applies a t-SVD based tensor nuclear norm to it, so that the imputation of each missing kernel entry exploits consistency information from the other views while being optimized jointly with kernel k-means clustering.
Key results
- On BDGP fea, LRKT-IMVC improves ACC over the state of the art by 5.8% to 13.0% across missing ratios 0.1 to 0.9.
- On HW2sources, LRKT-IMVC reaches 26.2% ACC and 18.1% NMI versus 17.9% and 10.7% for MKKM-IK-MKC.
- On UCI Digits, LRKT-IMVC reaches 77.4% ACC versus 54.9% for MKKM-IK-MKC, and it degrades less as the missing ratio grows.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the low-rank kernel tensor constraint to kernel imputation and clustering | LRKT-IMVC · BDGP fea | ACC: 37.6(3.1) -> 47.0(1.1) | Helped |
| add the low-rank kernel tensor constraint to kernel imputation and clustering | LRKT-IMVC · UCI Digits | ACC: 56.6(9.0) -> 77.4(1.8) | Helped |
| add the low-rank kernel tensor constraint to kernel imputation and clustering | LRKT-IMVC · Hdigit | ACC: 12.0(0.9) -> 19.5(4.7) | Helped |
| vary the low-rank constraint weight λ (ablation) | LRKT-IMVC · HW and CCV | ACC | Mixed |
Surprising findings
- With per-dataset tuning of lambda at each missing ratio, performance of LRKT-IMVC improves as the missing ratio increases, which the authors say is unusual for kernel-based IMVC methods.
Limitations
- Performance is sensitive to the trade-off parameter lambda, which the authors tune per missing ratio.
- Convergence is only to a local optimum, not the global optimum.
Experimental setup
Datasets: HW2sources UCI Digits HW Caltech101-20 BDGP fea CCV Hdigit
Benchmarks: HW2sources UCI Digits HW Caltech101-20 BDGP fea CCV Hdigit
Metrics: ACC NMI purity
Baselines: MKKM-IK-MKC MKKM-IK-ZF MKKM-IK-MF MKKM-IK-KNN MKKM+ZF MKKM+MF MKKM+KNN MKKM+LF BSV Concat DAIMC OPIMC
Builds on: MKKM-IK multiple kernel k-means tensor nuclear norm (t-TNN) ADMM
Closest papers by meaning
- KOALA: Kernel Coupling and Element Imputation Induced Multi-View Clustering (AAAI 2025 · similarity 0.92)
- EDISON: Enhanced Dictionary-Induced Tensorized Incomplete Multi-View Clustering with Gaussian Error Rank Minimization (ICML 2024 · similarity 0.91)
- Large-scale Multi-view Tensor Clustering with Implicit Linear Kernels (CVPR 2025 · similarity 0.90)
- SLR-MVTC: Smooth Low-Rank Multi-View Tensor Clustering (AAAI 2025 · similarity 0.89)
- Performance Gaps in Multi-view Clustering under the Nested Matrix-Tensor Model (ICLR 2024 · similarity 0.88)
- Incomplete Contrastive Multi-View Clustering with High-Confidence Guiding (AAAI 2024 · similarity 0.88)
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.