Cluster Specific Representation Learning

Mahalakshmi Sabanayagam, Omar Al-Dabooni, Pascal Esser · 2024

Read Cluster Specific Representation Learning by Mahalakshmi Sabanayagam, Omar Al-Dabooni, Pascal Esser (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-03471.txt.

Description

Mahalakshmi Sabanayagam, Omar Al-Dabooni, and Pascal Esser propose a new rule for judging learned representations of data. A representation shrinks data into a short vector form for later tasks. Quality is usually judged by one task at a time, yet a form that suits one task may fail another.

Their rule is task free: when data hold natural clusters, the representation should be specific to each cluster. A meta algorithm learns cluster specific forms and cluster assignments at the same time. It plugs into autoencoders, variational models, contrastive models, and Boltzmann machines. Tests show clearer cluster structure and better results on denoising and clustering, at a small cost in time and size.

Catalog record

CREATORMahalakshmi Sabanayagam, Omar Al-Dabooni, Pascal Esser
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE42 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-03471

Browse the full library · About the Tower of Babel