Research map / Learning theory and algorithms / Clustering
Deep Fair Multi-View Clustering with Attention KAN
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 DFMVC-AKAN, a deep fair multi-view clustering method that combines attention-augmented Kolmogorov-Arnold Network encoders with contrastive view consistency and a group-balanced distribution alignment loss to reduce sensitive-attribute dominance in clusters.
Problem
Multi-view clustering often lets sensitive attributes (e.g., gender, marital status) dominate cluster assignments, producing unfair results, and MLP/CNN encoders struggle to model complex nonlinear inter-view relationships.
Approach
Replaces MLP/CNN encoders with attention-augmented KAN layers (hybrid SE plus multi-head attention) and adds a fairness loss that aligns soft cluster assignments to a group-normalized target distribution so that each protected subgroup is balanced across clusters, combined with a cross-view contrastive clustering loss.
Key results
- On Bank Marketing, the method reaches NMI 80.46% versus 54.62% for the best baseline DFMVC.
- On Credit Card, NMI is 99.98% versus 93.93% for DFMVC.
- Ablation: removing the fairness loss lowers BAL on Bank Marketing from 42.52 to 41.59.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use DFMVC-AKAN instead of DFMVC | DFMVC-AKAN · Banking Market | NMI (%): 54.62 ± 1.25 -> 80.46 ± 0.51 | Helped |
| use DFMVC-AKAN instead of DFMVC | DFMVC-AKAN · Banking Market | BAL (%): 42.16 ± 0.82 -> 42.52 ± 0.32 | Helped |
| use DFMVC-AKAN instead of DFMVC | DFMVC-AKAN · Zafar | NMI (%): 93.93 ± 0.36 -> 99.98 ± 0.02 | Helped |
| use DFMVC-AKAN instead of DFMVC | DFMVC-AKAN · Zafar | BAL (%): 29.08 ± 0.28 -> 29.39 ± 0.14 | Helped |
| use DFMVC-AKAN instead of DFMVC | DFMVC-AKAN · Credit Card | NMI (%): 35.13 ± 0.62 -> 34.84 ± 0.21 | Hurt |
| use DFMVC-AKAN instead of DFMVC | DFMVC-AKAN · Credit Card | BAL (%): 39.71 ± 0.38 -> 39.98 ± 0.11 | Helped |
| use DFMVC-AKAN instead of DFMVC | DFMVC-AKAN · Law School | NMI (%): 24.24 ± 0.32 -> 23.64 ± 0.28 | Hurt |
| use DFMVC-AKAN instead of DFMVC | DFMVC-AKAN · Law School | BAL (%): 45.66 ± 0.38 -> 46.01 ± 0.04 | Helped |
| add the fairness learning module (ablation) | DFMVC-AKAN · Banking Market, Zafar, Credit Card, Law School | NMI: 76.36 -> 80.46; 95.23 -> 99.98; 24.69 -> 34.84; 21.44 -> 23.64 | Helped |
| add the fairness learning module (ablation) | DFMVC-AKAN · Banking Market, Zafar, Credit Card, Law School | BAL: 41.59 -> 42.52; 28.60 -> 29.39; 38.06 -> 39.98; 44.29 -> 46.01 | Helped |
| add the semantic contrastive learning module (ablation) | DFMVC-AKAN · Banking Market, Zafar, Credit Card, Law School | NMI: 59.73 -> 80.46; 86.66 -> 99.98; 19.98 -> 34.84; 17.85 -> 23.64 | Helped |
| add the semantic contrastive learning module (ablation) | DFMVC-AKAN · Banking Market, Zafar, Credit Card, Law School | BAL: 42.25 -> 42.52; 28.91 -> 29.39; 39.75 -> 39.98; 45.87 -> 46.01 | Helped |
Surprising findings
- Hyperparameter analysis reports Bank Marketing NMI around 62% for lambda > 1, indicating sensitivity to the reconstruction weight.
Limitations
- Results are reported on only four tabular datasets with two synthetic views generated by nonlinear functions.
- Sensitivity to hyperparameters (lambda, beta) is high, especially on Credit Card.
Experimental setup
Datasets: Bank Marketing Zafar Credit Card Law School
Benchmarks: Bank Marketing Zafar Credit Card Law School
Metrics: NMI BAL
Baselines: K-means DEC CC MvDSCN DCP APADC MFLVC DFMVC Fair-MVC
Builds on: DFMVC DEC Squeeze-and-Excitation Kolmogorov-Arnold network
Closest papers by meaning
- EASEMVC:Efficient Dual Selection Mechanism for Deep Multi-View Clustering (CVPR 2025 · similarity 0.89)
- Medusa: A Multi-Scale High-order Contrastive Dual-Diffusion Approach for Multi-View Clustering (CVPR 2025 · similarity 0.89)
- Enhanced then Progressive Fusion with View Graph for Multi-View Clustering (CVPR 2025 · similarity 0.88)
- A Non-parametric Graph Clustering Framework for Multi-View Data (AAAI 2024 · similarity 0.88)
- Adaptive Feature Imputation with Latent Graph for Deep Incomplete Multi-View Clustering (AAAI 2024 · similarity 0.87)
- Incomplete Contrastive Multi-View Clustering with High-Confidence Guiding (AAAI 2024 · similarity 0.87)
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.