Self-supervised non-dominated sorted model for co-clustering.

Area of Science:

• Data Science
• Machine Learning
• Computational Biology

Background:

• Co-clustering analyzes row and column structures and their inter-relationships, offering more insight than traditional methods.
• Co-clustering is inherently multi-objective, aiming to cluster samples and features while uncovering sample-feature relationships.
• Current co-clustering approaches often use single-objective optimization and neglect inherent data information.

Purpose of the Study:

• To propose a novel self-supervised non-dominated sorted model for co-clustering (SNSC).
• To address the multi-objective nature of co-clustering and incorporate supervised information from original data.
• To improve co-clustering efficiency and avoid local optima through heuristic and random initialization.

Main Methods:

• Developed a multi-objective function group for the SNSC model, including four objectives acting on data and similarity matrices.
• Implemented a hybrid initialization strategy combining heuristic self-supervised and random methods.
• Designed a genetic algorithm-based approach for the SNSC model, including theoretical support and complexity analysis.

Main Results:

• The SNSC model aligns with the multi-objective nature of co-clustering and utilizes inherent supervised information.
• The hybrid initialization enhances model efficiency and reduces convergence to local optima.
• Experimental results on 12 datasets demonstrate significant advantages of the SNSC algorithm over 5 comparison algorithms.

Conclusions:

• The proposed SNSC model effectively handles the multi-objective nature of co-clustering tasks.
• The self-supervised approach leverages data's inherent information without requiring external labels.
• SNSC demonstrates superior performance in co-clustering tasks, offering a promising advancement in data analysis.