Task-augmented cross-view imputation network for partial multi-view incomplete multi-label classification.

Area of Science:

• Computer Science
• Machine Learning
• Data Science

Background:

• Multi-view multi-label learning faces challenges with incomplete training data and missing features.
• Incomplete data hinders comprehensive sample understanding and accurate classification.

Purpose of the Study:

• To propose a novel network, TACVI-Net, for handling partial multi-view incomplete multi-label classification.
• To effectively recover missing views and improve classification performance.

Main Methods:

• A two-stage network, TACVI-Net, is developed to derive task-relevant features for imputing missing views.
• Stage 1: Information bottleneck theory and view-specific encoder-classifiers extract discriminative representations.
• Stage 2: An autoencoder-based multi-view reconstruction network augments features and imputes missing data.

Main Results:

• TACVI-Net demonstrates superior performance compared to existing state-of-the-art methods.
• Experiments on five diverse datasets validate the effectiveness of the proposed approach.

Conclusions:

• TACVI-Net successfully addresses the challenge of incomplete multi-view data in multi-label classification.
• The proposed imputation strategy significantly enhances classification accuracy by recovering crucial missing information.