Thetan Berserker: Fast and Stochastic Distance-based Clustering

Eleftherios Garyfallidis, Jong Sung Park, Sai Prajwal Reddy, Serge Koudoro

Keywords: clustering

TL;DR: Fast distance based clustering with a range of applications

Abstract: Clustering is a challenging NP-hard problem. Polynomial approximations are of paramount importance for identifying intriguing hidden representations of data at reasonable execution times. In this work, we propose a novel clustering algorithm called Thetan Berserker (TB). TB is a centroid-based clustering method controlled by a single distance parameter. TB revitalizes an old family of sequential algorithms which are adored for their speed but are known to be order sensitive. In addition, TB enables widely used algorithms such as KMeans and DBSCAN by improving their initial conditions. Theoretical aspects are provided in detail along with extensive comparisons and benchmarks. Examples of real world applications are provided using publicly available data of different dimensionalities. A wide range of performance boosts in clustering accuracy, memory usage, and runtime are reported. By dramatically reducing clustering ambiguities, while staying at incredibly low complexity, TB creates a new standard for clustering.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 3198