Cross-view discrepancy-driven dynamic weighting for missing view completion in incomplete multi-view clustering

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Incomplete Multi-view Clustering (IMVC) seeks shared structures across datasets with missing data.
• Existing methods often fail to account for noise introduced during data imputation, hindering performance.
• Noise in recovered data is a significant challenge in IMVC.

Purpose of the Study:

• To propose a novel dynamically weighted view completion method for IMVC.
• To enhance both data recovery quality and clustering performance by mitigating noise.
• To leverage cross-view discrepancy information for improved IMVC.

Main Methods:

• Employed cross-view contrastive learning to capture cross-view consistency and measure discrepancies.
• Developed a weight matrix using learned consistency features to assess recovered data quality.
• Implemented an iterative process optimizing view completion and discrepancy learning with a weighted recovery loss.

Main Results:

• The proposed method effectively reduces noise in imputed data.
• Demonstrated superior performance in both missing view completion and clustering accuracy compared to state-of-the-art methods.
• Experimental results on benchmark datasets validate the effectiveness of the dynamic weighting strategy.

Conclusions:

• The novel dynamically weighted view completion method significantly improves IMVC.
• Leveraging cross-view discrepancy information is crucial for robust data recovery and clustering.
• The proposed approach offers a promising solution for handling noise in incomplete multi-view data.