Research map / Learning theory and algorithms / Clustering
Tensorized Label Learning Based Fast Fuzzy 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 TLLFFC, a parameter-free anchor-based multi-view fuzzy clustering model that uses a balanced ℓ2,1 regularizer and tensor Schatten p-norm on stacked soft label matrices, optimized with ALM.
Problem
Multi-view graph clustering is costly for large-scale data, fuzzy clustering methods need extra regularization hyper-parameters to avoid trivial solutions, and view-specific label assignments are often inconsistent across views.
Approach
A hyper-parameter-free balanced regularizer (ℓ2,1 norm of H^T, with the anchor-graph label transmission F = B^T H) avoids trivial solutions and yields fuzzy memberships, while a tensor Schatten p-norm over the stacked view label matrices aligns labels across views.
Key results
- TLLFFC reaches ACC 1.000 on MSRC-v5 and MNIST4 and 0.999 on HandWritten, versus 0.990 for the best baseline Orth-NTF on MSRC.
- On Scene15, TLLFFC achieves ACC 0.840 versus 0.758 for Orth-NTF.
- On Reuters (18,758 samples), TLLFFC achieves ACC 0.808 versus 0.694 for Orth-NTF, while several methods run out of memory.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use TLLFFC instead of Orth-NTF | Multi-view clustering · MSRC | ACC: 0.990 -> 1 | Helped |
| use TLLFFC instead of Orth-NTF | Multi-view clustering · HandWritten | ACC: 0.985 -> 0.999 | Helped |
| use TLLFFC instead of TLL-AG | Multi-view clustering · MNIST | ACC: 0.981 -> 1 | Helped |
| use TLLFFC instead of Orth-NTF | Multi-view clustering · Scene | ACC: 0.758 -> 0.840 | Helped |
| include tensor Schatten p-norm regularization (ablation) | TLLFFC · MSRC | ACC: 0.742 -> 1 | Helped |
| include tensor Schatten p-norm regularization (ablation) | TLLFFC · MSRC | NMI: 0.654 -> 1 | Helped |
| include tensor Schatten p-norm regularization (ablation) | TLLFFC · MSRC | Purity: 0.776 -> 1 | Helped |
| include tensor Schatten p-norm regularization (ablation) | TLLFFC · HandWritten | ACC: 0.487 -> 0.999 | Helped |
| include tensor Schatten p-norm regularization (ablation) | TLLFFC · HandWritten | NMI: 0.578 -> 0.996 | Helped |
| include tensor Schatten p-norm regularization (ablation) | TLLFFC · MNIST | ACC: 0.612 -> 1 | Helped |
| include tensor Schatten p-norm regularization (ablation) | TLLFFC · Scene | ACC: 0.470 -> 0.840 | Helped |
| include tensor Schatten p-norm regularization (ablation) | TLLFFC · Scene | NMI: 0.502 -> 0.874 | Helped |
Surprising findings
- Ablation without the Schatten p-norm term drops ACC substantially (e.g., 0.487 vs 0.999 on HandWritten).
Limitations
- Performance depends on anchor graph quality, and results do not improve monotonically with anchor rate.
- The Schatten p-norm hyper-parameter lambda significantly affects results and must still be tuned.
- Method relies on MATLAB implementation with hardware details given only as Windows server with 128 GB RAM.
Experimental setup
Datasets: MSRC-v5 HandWritten MNIST4 Scene15 Reuters
Benchmarks: MSRC-v5 HandWritten MNIST4 Scene15 Reuters
Metrics: ACC NMI Purity
Baselines: CSMSC GMC ETLMSC LMVSC FMCNOF SFMC FPMVS-CAG Orth-NTF OMVFC-LICAG TLL-AG
Builds on: anchor graph multi-view clustering tensor Schatten p-norm label transmission iteratively re-weighted algorithm
Closest papers by meaning
- Tensorized Label Learning on Anchor Graph (AAAI 2024 · similarity 0.89)
- Large-scale Multi-view Tensor Clustering with Implicit Linear Kernels (CVPR 2025 · similarity 0.87)
- Embedded Feature Selection on Graph-Based Multi-View Clustering (AAAI 2024 · similarity 0.86)
- S2MVTC: a Simple yet Efficient Scalable Multi-View Tensor Clustering (CVPR 2024 · similarity 0.86)
- Fast Incomplete Multi-view Clustering by Flexible Anchor Learning (ICML 2025 · similarity 0.85)
- Local Graph Clustering with Noisy Labels (ICLR 2024 · similarity 0.85)
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.