Multi-view clustering improves consistency using adversarial learning
Adversarial Consistency-Guided Representation Learning for Multi-view Clustering
Machine Learning
Summary
Sometimes computers look at data from different views or types and try to put related things together, called clustering. But when learning shared features across these views, the computer might still remember which view a piece came from, causing confusion. The authors propose a method to hide this view identity by using a special process that tricks the model into ignoring view-specific clues. This helps the computer focus on what’s truly common across views, improving how well it groups related data. Their experiments show this approach works better than others on several test datasets.
What this means in practice
- •For data engineers: Improve data grouping accuracy in systems combining diverse data types by reducing confusion from view-specific details.
- •For business intelligence teams: Create more reliable customer or product segments by integrating different data sources while minimizing inconsistencies.
Authors
Yuchen Lin, Kunpeng Xu, Ying Fang, Lifei Chen
Abstract
Multi-view clustering aims to capture cross-view consistency while exploiting view-specific information. However, shared representations learned to capture cross-view consistency may still retain view-identifying information, potentially compromising the consistency of cross-view clustering structures. To address this issue, we propose ACGRL, an adversarial consistency-guided representation learning framework for multi-view clustering. ACGRL employs a gradient-reversal view discriminator to reduce view identifiability and obtain invariant reference representations. These representations are then frozen to provide fixed references for disentangling view-specific information from cross-view common information in the subsequent learning stage. The fixed reference representations are concatenated with the learned view-specific representations for reconstruction and clustering, with cross-view cluster alignment encouraging consistent clustering assignments. Experiments on four benchmark datasets demonstrate the superior clustering performance of ACGRL compared with representative multi-view clustering methods.