Research map / Label-efficient and robust learning / Recognition and long-tail
Unsupervised Kernel-based Multi-view Feature Selection with Robust Self-representation and Binary Hashing
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 UKMFS, an unsupervised kernel-based multi-view feature selection model combining RBF kernel pre-processing, low-rank multi-view consistent graph learning, and binary hashing pseudo-labels with an adaptive robust norm for feature selection.
Problem
Unsupervised multi-view feature selection on high-dimensional, possibly nonlinear multi-view data, where separate graph learning and feature selection lose semantic information and are sensitive to noise and outliers.
Approach
Jointly learns a consistent low-rank affinity graph across views via self-representation on RBF-kernel-mapped features, generates binary hash pseudo-labels from the graph Laplacian during spectral embedding, and uses an adaptive loss between Frobenius and l2,1 norms for feature selection, all optimized iteratively in one framework.
Key results
- On Caltech101 at 25% features, UKMFS reaches 70.62% ACC versus 70.42% for JMVFG (the best baseline on that row).
- On HW at 20% features, UKMFS achieves 85.52% ACC, the best among compared methods.
- Ablation shows removing the non-linear kernel process lowers performance relative to full UKMFS (20% features).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add nonlinear RBF kernel preprocessing (ablation) | UKMFS · Multi-view datasets | clustering ACC and NMI | Helped |
| Remove adaptive weight learning and the low-rank constraint on view-specific graphs (ablation) | UKMFS · Multi-view datasets | clustering ACC and NMI | Hurt |
| Replace learnable binary labels with a single-label approach (ablation) | UKMFS · Multi-view datasets | clustering ACC and NMI | Hurt |
Limitations
- Authors note future work is needed to reduce time complexity and improve efficiency.
- Parameter tuning via grid search over several hyperparameters (alpha, beta, gamma, hash length, anchor count).
Experimental setup
Datasets: Caltech101 HW MSRCV1 Scene ORL WebKB
Benchmarks: Caltech101 HW MSRCV1 Scene ORL WebKB
Metrics: ACC NMI
Baselines: UAFSBH ACSL NSGL AMFS JMVFG SCMvFS
Builds on: UAFSBH low-rank representation adaptive loss for robust feature selection
Closest papers by meaning
- EASEMVC:Efficient Dual Selection Mechanism for Deep Multi-View Clustering (CVPR 2025 · similarity 0.87)
- Embedded Feature Selection on Graph-Based Multi-View Clustering (AAAI 2024 · similarity 0.86)
- Incomplete and Unpaired Multi-View Graph Clustering with Cross-View Feature Fusion (AAAI 2025 · similarity 0.86)
- TIME-FS: Joint Learning of Tensorial Incomplete Multi-View Unsupervised Feature Selection and Missing-View Imputation (AAAI 2025 · similarity 0.85)
- TPCH: Tensor-interacted Projection and Cooperative Hashing for Multi-view Clustering (AAAI 2025 · similarity 0.85)
- Double-Layer Hybrid-Label Identification Feature Selection for Multi-View Multi-Label Learning (AAAI 2024 · similarity 0.84)
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.