Synergistic Prompting for Complementarity and Consistency in Incomplete Multi-View Clustering.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Incomplete multi-view clustering (IMVC) addresses partitioning unlabeled data with missing views.
• Existing deep IMVC methods struggle with cross-view complementarity and global semantic consistency.

Purpose of the Study:

• To propose SP-IMVC, a novel framework for IMVC that jointly models complementarity and consistency under view incompleteness.
• To address the limitations of existing IMVC methods in handling missing data and ensuring semantic coherence.

Main Methods:

• Introduced Synergistic Prompting (SP-IMVC) with two learnable prompts: Cross-View Complementary Prompt (CVCP) and Latent Anchor Prompt (LAP).
• CVCP aggregates auxiliary representations to enrich semantics and mitigate information loss.
• LAP uses a global anchor prompt pool for adaptive semantic priors, promoting consistent representations.

Main Results:

• SP-IMVC consistently outperforms 14 state-of-the-art IMVC approaches across six benchmarks.
• Demonstrated superior performance in scenarios with high missing-view ratios.
• Validated the effectiveness and robustness of the synergistic prompt-guided clustering framework.

Conclusions:

• SP-IMVC effectively models cross-view complementarity and global semantic consistency in incomplete multi-view data.
• The proposed framework offers a robust solution for IMVC, particularly under significant data missingness.
• The study validates the synergistic prompting approach for enhancing IMVC performance.