Compressibility Measures Complexity: Minimum Description Length Meets Singular Learning Theory

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Urdshals, Einar, Lau, Edmund, Hoogland, Jesse, van Wingerden, Stan, Murfet, Daniel
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866917011465961472
author Urdshals, Einar
Lau, Edmund
Hoogland, Jesse
van Wingerden, Stan
Murfet, Daniel
author_facet Urdshals, Einar
Lau, Edmund
Hoogland, Jesse
van Wingerden, Stan
Murfet, Daniel
contents We study neural network compressibility by using singular learning theory to extend the minimum description length (MDL) principle to singular models like neural networks. Through extensive experiments on the Pythia suite with quantization, factorization, and other compression techniques, we find that complexity estimates based on the local learning coefficient (LLC) are closely, and in some cases, linearly correlated with compressibility. Our results provide a path toward rigorously evaluating the limits of model compression.
format Preprint
id arxiv_https___arxiv_org_abs_2510_12077
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Compressibility Measures Complexity: Minimum Description Length Meets Singular Learning Theory
Urdshals, Einar
Lau, Edmund
Hoogland, Jesse
van Wingerden, Stan
Murfet, Daniel
Machine Learning
We study neural network compressibility by using singular learning theory to extend the minimum description length (MDL) principle to singular models like neural networks. Through extensive experiments on the Pythia suite with quantization, factorization, and other compression techniques, we find that complexity estimates based on the local learning coefficient (LLC) are closely, and in some cases, linearly correlated with compressibility. Our results provide a path toward rigorously evaluating the limits of model compression.
title Compressibility Measures Complexity: Minimum Description Length Meets Singular Learning Theory
topic Machine Learning
url https://arxiv.org/abs/2510.12077