Research map / Learning theory and algorithms / Clustering

S2MVTC: a Simple yet Efficient Scalable Multi-View Tensor Clustering

CVPR 2024 · Learning theory and algorithms › Clustering · approach: Incomplete multi-view clustering · 21 citations (OpenAlex)

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 S2MVTC, a scalable anchor-based multi-view clustering method that stacks per-view embedding features into a rotated tensor and applies a tensor low-frequency approximation (TLFA) plus a consensus constraint to learn intra- and inter-view embedding correlations.

Problem

Large-scale multi-view clustering methods mainly seek consensus through global correlations between anchor graphs or projection matrices, and are costly or fail to exploit correlations among embedding features within and across views.

Approach

Directly learn intra-view graph similarity and inter-view semantic consistency on the embedding feature tensor: a tensor low-frequency approximation (TLFA) in the Fourier domain along the sample mode smooths embeddings, while a consensus term pulls per-view embeddings to their average.

Key results

• On CCV, S2MVTC reaches 54.79% ACC versus 45.54% for sMERA-MVC (sMERA-MVC row in Table 3 reports 0.4554 ACC), with 1.61 s CPU time versus 3.81 s.
• On AwA, S2MVTC achieves 62.88% ACC and a 77.58x speedup over k-means in CPU time (10.35 s).
• Ablation on six datasets shows removing the intra-view graph similarity (TLFA) module drops ACC sharply, e.g. CCV from 54.79% to 15.49%.

Measured effects reported in the paper

Surprising findings

• Removing the TLFA (intra-view graph similarity) component collapses accuracy far more than removing the inter-view consensus term, which only slightly reduces performance.

Limitations

• Hyperparameters (beta, lambda, L, M) are tuned by grid search per dataset, with sensitivity shown mainly on Caltech102.
• Maximum iterations fixed at T=7 with tau growing geometrically, limiting analysis of convergence beyond a theoretical argument.

Experimental setup

Datasets: CCV Caltech102 NUS-WIDE-OBJ AwA Cifar-10 YoutubeFace sel

Benchmarks: CCV Caltech102 NUS-WIDE-OBJ AwA Cifar-10 YoutubeFace sel

Metrics: ACC NMI Purity F-score PRE REC ARI CPU time

Baselines: k-means BMVC SFMC OPMC FPMVS-CAG FMVACC FastMICE FSMSC sMERA-MVC

Builds on: anchor graph multi-view clustering t-SVD tensor nuclear norm RBF kernel anchor graph

Closest papers by meaning

• SLR-MVTC: Smooth Low-Rank Multi-View Tensor Clustering (AAAI 2025 · similarity 0.92)
• Large-scale Multi-view Tensor Clustering with Implicit Linear Kernels (CVPR 2025 · similarity 0.91)
• From Dictionary to Tensor: A Scalable Multi-View Subspace Clustering Framework with Triple Information Enhancement (NeurIPS 2024 · similarity 0.91)
• Performance Gaps in Multi-view Clustering under the Nested Matrix-Tensor Model (ICLR 2024 · similarity 0.90)
• Scalable Federated One-Step Multi-View Clustering with Tensorized Regularization (AAAI 2025 · similarity 0.89)
• Fast Incomplete Multi-view Clustering with Adaptive Similarity Completion and Reconstruction (AAAI 2025 · similarity 0.89)