Ensemble Clustering using Semidefinite Programming with Applications.

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Ensemble clustering aggregates multiple clustering solutions to find a consensus.
• Existing methods often rely on voting strategies to measure agreement.
• There is a need for more effective agreement measures and optimization techniques.

Purpose of the Study:

• To develop a novel approach for ensemble clustering using 2D string encoding.
• To improve the aggregation of multiple clustering solutions.
• To explore new applications of ensemble clustering.

Main Methods:

• Developed a 2D string encoding to capture agreement in clustering ensembles.
• Constructed a non-linear objective function and transformed it into a 0-1 Semidefinite Program (SDP).
• Relaxed the SDP to a polynomial-time solvable form using convexification techniques.

Main Results:

• The 2D string encoding approach demonstrates improved agreement compared to voting strategies.
• Experimental results show enhanced performance on machine learning and synthetic datasets.
• The method is effective across various agreement measures.

Conclusions:

• The proposed 2D string encoding offers a superior method for ensemble clustering.
• This approach has potential applications in image segmentation and diffusion tensor imaging.
• The study advances ensemble clustering techniques and their practical utility.