Clustering in Transformed Feature Spaces by Analyzing Distinct Modes
MCML Authors
Collin Leiber
Dr.
* Former Member
Abstract
Collin Leiber
Dr.
* Former Member
Abstract
The growing availability of data demands clustering methods that can extract valuable information without requiring costly annotations, especially for large, high-dimensional datasets. This dissertation develops subspace and deep clustering approaches, leveraging methods like the Dip-test of unimodality and Minimum Description Length principle to identify and encode relevant features and clusters automatically, even in complex datasets. By incorporating these techniques into neural networks and refining them through a novel parameter-free approach, the research offers robust clustering tools that perform well without prior knowledge of the number of clusters, all implemented in the open-source package ClustPy. (Shortened).
BibTeXKey: Lei24