Parameter-free entropy-regularized multi-view clustering with hierarchical feature selection

2025-08-07

AI Summary

Multi-view clustering faces challenges in high-dimensional heterogeneous data, including automatic pattern discovery, redundant feature suppression, and consistent cross-view representation learning; existing methods rely heavily on manual hyperparameter tuning and lack principled integration mechanisms. To address these issues, we propose AAMVFCM-U, a parameter-free entropy-regularized framework. It introduces signal-to-noise-ratio-driven feature weighting and dual-layer entropy constraints to enable adaptive cross-view consensus learning and joint feature selection. Additionally, a hierarchical dimensionality reduction strategy is designed, employing adaptive thresholds to dynamically identify critical view combinations while ensuring algorithmic convergence. Extensive experiments on five benchmark datasets demonstrate that AAMVFCM-U significantly outperforms 15 state-of-the-art methods. It achieves up to 97% improvement in computational efficiency and compresses feature dimensions by 99.55%—retaining only 0.45% of original features—without any hyperparameter tuning.

Technology Category

• Machine Learning: Multi-instance/Multi-view Learning
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Multi-view clustering faces critical challenges in automatically discovering patterns across heterogeneous data while managing high-dimensional features and eliminating irrelevant information. Traditional approaches suffer from manual parameter tuning and lack principled cross-view integration mechanisms. This work introduces two complementary algorithms: AMVFCM-U and AAMVFCM-U, providing a unified parameter-free framework. Our approach replaces fuzzification parameters with entropy regularization terms that enforce adaptive cross-view consensus. The core innovation employs signal-to-noise ratio based regularization ($δ_j^h = frac{ar{x}_j^h}{(σ_j^h)^2}$) for principled feature weighting with convergence guarantees, coupled with dual-level entropy terms that automatically balance view and feature contributions. AAMVFCM-U extends this with hierarchical dimensionality reduction operating at feature and view levels through adaptive thresholding ($θ^{h^{(t)}} = frac{d_h^{(t)}}{n}$). Evaluation across five diverse benchmarks demonstrates superiority over 15 state-of-the-art methods. AAMVFCM-U achieves up to 97% computational efficiency gains, reduces dimensionality to 0.45% of original size, and automatically identifies critical view combinations for optimal pattern discovery.

Problem

• Automatically discovers patterns across heterogeneous multi-view data
• Manages high-dimensional features and eliminates irrelevant information
• Eliminates manual parameter tuning through unified parameter-free framework

Innovation

• Parameter-free entropy-regularized multi-view clustering framework
• Signal-to-noise ratio based feature weighting with convergence
• Hierarchical dimensionality reduction through adaptive thresholding mechanisms