Research map / Graph learning / Graph algorithms and clustering
Fast Incomplete Multi-view Clustering by Flexible Anchor Learning
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 FIML, a unified fast incomplete multi-view clustering method that jointly learns a shared anchor graph, flexible anchors, and cluster partition via alternating optimization.
Problem
Clustering incomplete multi-view data at large scale, where anchor learning, graph construction, and graph partition are usually handled as separate stages, limiting anchor quality and scalability.
Approach
Learns consensus anchor projections and a shared anchor graph with missing-sample indicators, and links the anchor graph to a symmetric nonnegative factorization with orthogonal centroid constraints so cluster assignments are obtained directly without a separate spectral clustering step.
Key results
- FIML reaches 78.84% ACC on ORL vs 76.30% for the best baseline FIMVC-VIA, and 99.50% ACC on Cifar100.
- On STL10 FIML achieves 78.30% ACC, 59.80% NMI, and 62.50% F1 with unified optimization versus 70.60% ACC in the separated-stage ablation on ORL.
- Running time on MNIST is 18.40 s for FIML versus 20.20 s for FIMVC-VIA and 552 s for IMVC-CBG; several baselines run out of memory on MNIST and Cifar100.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use a unified framework instead of separating graph construction, anchor learning, and graph partition (ablation) | FIML · ORL | ACC (%): 70.60±0.20 -> 78.84±0.50 | Helped |
| use a unified framework instead of separating graph construction, anchor learning, and graph partition (ablation) | FIML · NGs | ACC (%): 82.45±0.30 -> 91.40±0.40 | Helped |
| use a unified framework instead of separating graph construction, anchor learning, and graph partition (ablation) | FIML · WebKB | ACC (%): 82.00±0.70 -> 93.00±0.26 | Helped |
| use a unified framework instead of separating graph construction, anchor learning, and graph partition (ablation) | FIML · STL10 | ACC (%): 71.40±0.55 -> 78.30±0.60 | Helped |
| use a unified framework instead of separating graph construction, anchor learning, and graph partition (ablation) | FIML · MNIST | ACC (%): 84.60±0.00 -> 98.90±0.00 | Helped |
| use a unified framework instead of separating graph construction, anchor learning, and graph partition (ablation) | FIML · Cifar100 | ACC (%): 90.40±0.45 -> 99.50±0.26 | Helped |
Surprising findings
- Several baselines (MIC, MKKM-IK, DAIMC, APMC, PIC, etc.) run out of memory on MNIST and Cifar100, while FIML scales to them.
Limitations
- Relies on hyperparameters lambda and number of anchors m; sensitivity studied only on NGs and ORL.
- Some baselines are not compared on all datasets due to out-of-memory failures, limiting comparison breadth.
- Ablation compares only a separated-stage variant rather than individual component removal.
Experimental setup
Datasets: NGs WebKB ORL STL10 MNIST Cifar100
Benchmarks: NGs WebKB ORL STL10 MNIST Cifar100
Metrics: ACC NMI F1-score Purity running time
Baselines: BSV MIC MKKM-IK DAIMC APMC PIC EEIMVC V3H IMVC-CBG FIMVC-VIA
Builds on: anchor graph clustering symmetric nonnegative matrix factorization IMVC-CBG FIMVC-VIA
Closest papers by meaning
- Scalable One-Pass Incomplete Multi-View Clustering by Aligning Anchors (AAAI 2025 · similarity 0.94)
- Fast Incomplete Multi-view Clustering with Adaptive Similarity Completion and Reconstruction (AAAI 2025 · similarity 0.93)
- Alleviate Anchor-Shift: Explore Blind Spots with Cross-View Reconstruction for Incomplete Multi-View Clustering (NeurIPS 2024 · similarity 0.92)
- DVSAI: Diverse View-Shared Anchors Based Incomplete Multi-View Clustering (AAAI 2024 · similarity 0.92)
- Sample-Level Cross-View Similarity Learning for Incomplete Multi-View Clustering (AAAI 2024 · similarity 0.91)
- Learn from View Correlation: An Anchor Enhancement Strategy for Multi-view Clustering (CVPR 2024 · similarity 0.91)
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.