Information theoretic underpinning of self-supervised learning by clustering

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Kittler, Josef, Atito, Sara, Awais, Muhammad
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866916004486971392
author Kittler, Josef
Atito, Sara
Awais, Muhammad
author_facet Kittler, Josef
Atito, Sara
Awais, Muhammad
contents Self-supervised learning (SSL) is recognized as an essential tool for building foundation models for Artificial Intelligence applications. The advances in SSL have been made thanks to vigorous arguments about the principles of SSL and through extensive empirical research. The aim of this paper is to contribute to the development of the underpinning theory of SSL, focusing on the deep clustering approach. By analogy to supervised learning, we formulate SSL as K-L divergence optimization. The mode collapse is prevented by imposing an optimisation constraint on the teacher distribution. This leads to normalization using inverse cluster priors. We show that using Jensen inequality this normalization simplifies to the popular batch centering procedure. Distillation and centering are common {heuristics-based} practices in SSL, {but our work underpins them theoretically.} The theoretical model developed not only supports specific existing successful SSL methods, but also suggests directions for future investigations.
format Preprint
id arxiv_https___arxiv_org_abs_2605_11870
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Information theoretic underpinning of self-supervised learning by clustering
Kittler, Josef
Atito, Sara
Awais, Muhammad
Machine Learning
Information Theory
Self-supervised learning (SSL) is recognized as an essential tool for building foundation models for Artificial Intelligence applications. The advances in SSL have been made thanks to vigorous arguments about the principles of SSL and through extensive empirical research. The aim of this paper is to contribute to the development of the underpinning theory of SSL, focusing on the deep clustering approach. By analogy to supervised learning, we formulate SSL as K-L divergence optimization. The mode collapse is prevented by imposing an optimisation constraint on the teacher distribution. This leads to normalization using inverse cluster priors. We show that using Jensen inequality this normalization simplifies to the popular batch centering procedure. Distillation and centering are common {heuristics-based} practices in SSL, {but our work underpins them theoretically.} The theoretical model developed not only supports specific existing successful SSL methods, but also suggests directions for future investigations.
title Information theoretic underpinning of self-supervised learning by clustering
topic Machine Learning
Information Theory
url https://arxiv.org/abs/2605.11870