Research map / Learning theory and algorithms / Clustering
Alleviate Anchor-Shift: Explore Blind Spots with Cross-View Reconstruction 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 AIMC-CVR, an anchor-based incomplete multi-view clustering method that uses cross-view anchor graph learning and affine-combination reconstruction of missing samples to alleviate anchor shift.
Problem
Anchor-based incomplete multi-view clustering suffers from anchor shift caused by missing samples, which misaligns anchor graphs across views and degrades clustering quality on large-scale incomplete data.
Approach
Learns a complete anchor graph via symmetric cross-view projections so anchors are guided by other views, and reconstructs missing samples with affine (rather than convex) combinations of anchors, iterated with anchor learning, so reconstructions can lie outside the anchor convex hull.
Key results
- On MSRCV, WebKB, Wiki and Hdigit, AIMC-CVR achieves the highest ACC, e.g. 83.52% on MSRCV and 91.37% on WebKB, surpassing second-best by 3.65% and 7.89% respectively.
- On large-scale YTF10, YTF20 and MNIST, ACC improves over the second-best anchor-based method by 3.18%, 0.42% and 7.47%.
- Ablation: replacing affine with convex combination (v5) drops MSRCV ACC from 83.52% to 71.92%.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add cross-view anchor learning (ablation) | AIMC-CVR · MSRCV, WebKB, Wiki, Hdigit, YTF10, YTF20, MNIST | ACC(%): 67.80±0.92 -> 83.52±1.00 (MSRCV); 72.77±0.00 -> 91.37±0.00 (WebKB) | Helped |
| add affine combination-based reconstruction (ablation) | AIMC-CVR · MSRCV, WebKB, Wiki, Hdigit, YTF10, YTF20, MNIST | ACC(%): 71.95±0.75 -> 83.52±1.00 (MSRCV); 72.18±0.00 -> 91.37±0.00 (WebKB) | Helped |
| add sparsity regularization to the anchor graph (ablation) | AIMC-CVR · MSRCV, WebKB, Wiki, Hdigit, YTF10, YTF20, MNIST | ACC(%): 44.41±1.89 -> 83.52±1.00 (MSRCV); 87.47±0.00 -> 91.37±0.00 (WebKB) | Mixed |
| update anchors during optimization instead of keeping initialized anchors fixed (ablation) | AIMC-CVR · MSRCV, WebKB, Wiki, Hdigit, YTF10, YTF20, MNIST | ACC(%): 66.23±1.50 -> 83.52±1.00 (MSRCV); 27.66±0.46 -> 82.29±0.01 (Hdigit) | Helped |
| use affine rather than convex combination for reconstruction (ablation) | AIMC-CVR · MSRCV, WebKB, Wiki, Hdigit, YTF10, YTF20, MNIST | ACC(%): 71.92±0.73 -> 83.52±1.00 (MSRCV); 78.96±0.57 -> 83.73±1.29 (YTF10) | Helped |
Surprising findings
- Removing the sparsity regularization (beta=0) causes a large drop, e.g. MSRCV ACC falls to 44.41%.
Limitations
- Difficulty handling high-dimensional data due to O(d_p^3) projection cost and O(d_p d_q) space.
- Does not assign different confidence levels to relationships between views in cross-view learning.
- Convergence only guaranteed to a local minimum for a non-convex objective.
Experimental setup
Datasets: MSRCV WebKB Wiki Hdigit YTF10 YTF20 MNIST
Benchmarks: MSRCV WebKB Wiki Hdigit YTF10 YTF20 MNIST
Metrics: ACC NMI Purity Fscore
Baselines: DAIMC UEAF EEIMVC FLSD V3H IMVC-CBG SCBGL DVSAI
Builds on: anchor graph multi-view clustering IMVC-CBG k-means on singular vectors of anchor graph
Closest papers by meaning
- Scalable One-Pass Incomplete Multi-View Clustering by Aligning Anchors (AAAI 2025 · similarity 0.92)
- Fast Incomplete Multi-view Clustering by Flexible Anchor Learning (ICML 2025 · similarity 0.92)
- Learn from View Correlation: An Anchor Enhancement Strategy for Multi-view Clustering (CVPR 2024 · similarity 0.92)
- Fast Incomplete Multi-view Clustering with Adaptive Similarity Completion and Reconstruction (AAAI 2025 · similarity 0.90)
- Max-Mahalanobis Anchors Guidance for Multi-View Clustering (AAAI 2025 · similarity 0.90)
- Sample-Level Cross-View Similarity Learning for Incomplete Multi-View Clustering (AAAI 2024 · similarity 0.90)
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.