Transformed K-means Clustering

TL;DR

This paper introduces a novel clustering framework that integrates transform learning with K-means, optimizing the combined objective to enhance clustering performance, especially demonstrated on document datasets.

Contribution

It presents a new method that embeds K-means loss into transform learning and solves the joint problem with ADMM, improving clustering results over existing methods.

Findings

Improved document clustering accuracy

Outperforms state-of-the-art methods

Effective integration of transform learning with K-means

Abstract

In this work we propose a clustering framework based on the paradigm of transform learning. In simple terms the representation from transform learning is used for K-means clustering; however, the problem is not solved in such a na\"ive piecemeal fashion. The K-means clustering loss is embedded into the transform learning framework and the joint problem is solved using the alternating direction method of multipliers. Results on document clustering show that our proposed approach improves over the state-of-the-art.