View-Adaptive Multi-Granularity Anchor Learning for Multi-View Clustering.

Area of Science:

• Data Mining
• Machine Learning
• Computer Vision

Background:

• Multi-view clustering (MVC) enhances data analysis by integrating information from multiple sources.
• Current MVC methods often use single-granularity anchors, leading to suboptimal information mining.
• Aggregating anchors across varying scales for shared clustering remains a challenge.

Purpose of the Study:

• To propose a novel MVC method, View-adaptive Multi-granularity Anchor Learning (VMAL).
• To address limitations of single-granularity anchor learning in existing MVC techniques.
• To achieve improved clustering accuracy and efficiency through adaptive anchor management.

Main Methods:

• VMAL employs view-adaptive anchor pruning and view-shared sample clustering, optimized jointly.
• It dynamically adjusts the number of anchors per view using sample reconstruction error.
• A mapping-aggregation message passing strategy transfers anchor cluster information to samples.

Main Results:

• VMAL successfully obtains discrete sample cluster distributions without post-processing.
• Experimental results show VMAL outperforms state-of-the-art methods on multiple datasets.
• The method demonstrates superior clustering performance and efficiency.

Conclusions:

• VMAL offers a significant advancement in multi-view clustering.
• The view-adaptive, multi-granularity anchor approach effectively overcomes previous limitations.
• This method provides a robust framework for enhanced multi-view data analysis.