A game-inspired algorithm for marginal and global clustering

Miguel de Carvalho, Gabriel Martos, Andrej Svetlošák

Research output: Contribution to journalArticlepeer-review

Abstract

Clustering is an unsupervised learning approach for the task of partitioning data into meaningful subsets. The huge literature on cluster analysis is difficult to survey in a few sentences, but a concise description of well-known approaches is offered by [1–3]. True clusters, can be regarded in a number of different ways, each offering unique insights into the underlying patterns and relationships [4].
Examples of mainstream methods for clustering data include model-based (i.e., via mixture models), similarity-based (i.e., via K-means and K-medoids), and hierarchical clustering (i.e., clustering via dendograms). Model-based clustering is a fast-evolving and intradisciplinary research topic as can be seen from the recent papers of [5, 6], the survey papers of [7–9], and the Handbook on Mixture Analysis [10].
Despite decades of development in similarity-based clustering, K-means algorithms also remains in widespread use, with refined versions continually emerging [11, 12].
In this paper we propose a novel game-inspired method for cluster analysis that lies at the
interface of model-based and similarity-based clustering. The proposed approach aims to benefit from the flexibility and soundness of clustering via mixture models, while attempting to mitigate Pitfalls 1 and 2 below. In words, Pitfall 1 refers to an often overlooked aspect of model-based clustering: it usually leads to the same number of clusters for each margin—an assumption that may not align with practical applications. Pitfall 2 constrains the applicability of model-based clustering in high-dimensional data settings.
Original languageEnglish
Article number111158
JournalPattern Recognition
Volume160
Early online date22 Nov 2024
DOIs
Publication statusE-pub ahead of print - 22 Nov 2024

Fingerprint

Dive into the research topics of 'A game-inspired algorithm for marginal and global clustering'. Together they form a unique fingerprint.

Cite this