# Multi-view Semantic Consistency based Information Bottleneck for Clustering

**Authors:** Wenbiao Yan, Jihua Zhu, Yiyang Zhou, Yifei Wang, Qinghai Zheng  
**arXiv:** [2303.0002](https://arxiv.org/abs/2303.0002)  
**Date:** 2023-02-28  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.0002v1/

## Key points

- The paper proposes a novel approach to multi-view clustering that leverages semantic consistency.
- Existing methods focus on learning a fused representation matrix, ignoring private information and noise.
- MSCIB conducts an alignment operation of multiple views in the semantic space to achieve valuable consistent information from multi-view data.
- Learned semantic consistency improves the information bottleneck's ability to distinguish consistent information more accurately and learn a unified feature representation with more discriminative consistent information for clustering.
- MSCIB outperforms state-of-the-art methods on various types of multi-view datasets.
- The approach has significant implications in real-world applications where multiple sources of data are available, such as medical diagnosis or social network analysis.
- By leveraging semantic consistency across multiple views, MSCIB can provide more accurate and robust clustering results while accounting for private information and noise in each view.
- The paper presents a promising direction for future research in multi-view clustering by incorporating semantic consistency into the learning process.

## Summary

The paper titled "Multi-view Semantic Consistency based Information Bottleneck for Clustering" proposes a novel approach to multi-view clustering that leverages semantic consistency to improve the learning process of information bottleneck for different views. While most existing methods focus on learning a fused representation matrix, they ignore the influence of private information and noise. To address this limitation, the authors introduce MSCIB, which conducts an alignment operation of multiple views in the semantic space and jointly achieves valuable consistent information from multi-view data. This learned semantic consistency improves the information bottleneck's ability to distinguish consistent information more accurately and learn a unified feature representation with more discriminative consistent information for clustering. The proposed method is evaluated on various types of multi-view datasets, and experimental results show that MSCIB outperforms state-of-the-art methods. The authors' approach has significant implications in real-world applications where multiple sources of data are available, such as medical diagnosis or social network analysis. By leveraging semantic consistency across multiple views, MSCIB can provide more accurate and robust clustering results while accounting for private information and noise in each view. Overall, this paper presents a promising direction for future research in multi-view clustering by incorporating semantic consistency into the learning process.