Advances in Correlation Clustering
MCML Authors
Daniyal Kazempour
Dr.
* Former Member
Abstract
Daniyal Kazempour
Dr.
* Former Member
Abstract
This thesis addresses key challenges in correlation clustering, particularly in high-dimensional datasets, by developing novel methods to evaluate and improve clustering algorithms. The first contribution focuses on defining and deriving internal evaluation criteria for correlation clustering, proposing a new cost function to assess cluster quality based on commonalities among existing algorithms. The second part introduces two innovative strategies for detecting regions of interest (ROIs) in Hough space, improving the robustness of the Hough transform algorithm, and extending it to handle quadratic and periodic correlated clusters. Finally, the thesis explores unifying local and global correlation clustering views and enhancing the resilience of these methods to outliers. (Shortened.)
BibTeXKey: Kaz22