Research map / Graph learning / Graph algorithms and clustering
Highly Efficient Rotation-Invariant Spectral Embedding for Scalable 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 RISE, a bipartite-graph spectral embedding method for incomplete multi-view clustering that recovers a consensus embedding with second-order rotation invariance and solves it with a fast SVD-based alternating optimization of linear complexity.
Problem
Incomplete multi-view clustering suffers from missing views/instances, high computational cost of graph-based completion, and spectral embedding rotational mismatch across views with incomplete data.
Approach
Learn view-specific spectral embeddings from incomplete anchor-based bipartite graphs and jointly recover a complete consensus embedding via a second-order (YY^T vs. F F^T) rotation-invariant term; the alternating updates reduce to SVDs of concatenated small matrices, giving linear complexity in n.
Key results
- On Prokaryotic, RISE reaches 75.13 ACC versus 60.42 for the second-best SIMVC-SA (averaged over nine missing ratios).
- On YoutubeFace, RISE runs in 899.54 s versus 3209.01 s for SIMVC-SA, and is fastest on all datasets except Caltech101-7.
- RISE is reported to be about 10x faster than the first-order completion strategy on Prokaryotic and about 100x faster on Caltech101-7.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use RISE instead of SIMVC-SA | incomplete multi-view clustering · Prokaryotic | ACC: +14.71% | Helped |
| Use RISE instead of GSRIMC | incomplete multi-view clustering · WebKB | ACC: +9.42% | Helped |
| Use RISE instead of HCLS-CGL | incomplete multi-view clustering · Caltech101-7 | ACC: +15.88% | Helped |
| Use RISE instead of sFSR-IMVC | incomplete multi-view clustering · Wikipedia | ACC: +1.42% | Helped |
| Use RISE instead of SIMVC-SA | incomplete multi-view clustering · CIFAR10 | ACC: +0.5% | Helped |
| Use RISE instead of SIMVC-SA | incomplete multi-view clustering · FMNIST | ACC: +1.5% | Helped |
| Use RISE instead of IMVC-CBG | incomplete multi-view clustering · YoutubeFace | ACC: +9.84% | Helped |
| Use k-means anchors instead of DAS anchors (ablation) | RISE · Prokaryotic | ACC: over 15% | Helped |
| Use k-means anchors instead of DAS anchors (ablation) | RISE · Caltech101-7 | ACC: around 10% | Helped |
| Use second-order rotation-invariant completion instead of first-order completion (ablation) | RISE · Prokaryotic and WebKB | ACC and NMI | Helped |
| Use second-order rotation-invariant completion instead of first-order completion (ablation) | RISE · Prokaryotic | runtime: nearly 10 times lower | Helped |
| Use second-order rotation-invariant completion instead of first-order completion (ablation) | RISE · Caltech101-7 | runtime: around 100 times lower | Helped |
Surprising findings
- On Wikipedia, RISE's NMI (35.78) is lower than several baselines such as SIMVC-SA (39.18) and IMVC-CBG (37.96).
Limitations
- Performance depends on hyperparameters beta, anchor number m, and embedding dimension k, which require tuning.
- Authors note that developing a parameter-free model needs more attention.
- Full-similarity-graph variant cannot use the simplified SVD update, so bipartite graphs are required for efficiency.
Experimental setup
Datasets: Prokaryotic WebKB Caltech101-7 Wikipedia CIFAR10 FMNIST YoutubeFace
Benchmarks: Prokaryotic WebKB Caltech101-7 Wikipedia CIFAR10 FMNIST YoutubeFace
Metrics: ACC NMI Purity running time (s)
Baselines: UEAF PIMVC HCLS-CGL GSRIMC IMVTSC-MVI DAIMC sFSR-IMVC PSIMVC-PG IMVC-CBG SIMVC-SA
Builds on: anchor graph / bipartite graph multi-view clustering spectral clustering k-means anchor selection
Closest papers by meaning
- DVSAI: Diverse View-Shared Anchors Based Incomplete Multi-View Clustering (AAAI 2024 · similarity 0.89)
- Fast Incomplete Multi-view Clustering with Adaptive Similarity Completion and Reconstruction (AAAI 2025 · similarity 0.89)
- From Spectrum-free towards Baseline-view-free: Double-track Proximity Driven Multi-view Clustering (ICML 2025 · similarity 0.88)
- Sample-Level Cross-View Similarity Learning for Incomplete Multi-View Clustering (AAAI 2024 · similarity 0.88)
- Bifurcate then Alienate: Incomplete Multi-view Clustering via Coupled Distribution Learning with Linear Overhead (ICML 2025 · 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.