A kinetic-based regularization method for data science applications

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Ganguly, Abhisek, Gabbana, Alessandro, Rao, Vybhav, Succi, Sauro, Ansumali, Santosh
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866912542765350912
author Ganguly, Abhisek
Gabbana, Alessandro
Rao, Vybhav
Succi, Sauro
Ansumali, Santosh
author_facet Ganguly, Abhisek
Gabbana, Alessandro
Rao, Vybhav
Succi, Sauro
Ansumali, Santosh
contents We propose a physics-based regularization technique for function learning, inspired by statistical mechanics. By drawing an analogy between optimizing the parameters of an interpolator and minimizing the energy of a system, we introduce corrections that impose constraints on the lower-order moments of the data distribution. This minimizes the discrepancy between the discrete and continuum representations of the data, in turn allowing to access more favorable energy landscapes, thus improving the accuracy of the interpolator. Our approach improves performance in both interpolation and regression tasks, even in high-dimensional spaces. Unlike traditional methods, it does not require empirical parameter tuning, making it particularly effective for handling noisy data. We also show that thanks to its local nature, the method offers computational and memory efficiency advantages over Radial Basis Function interpolators, especially for large datasets.
format Preprint
id arxiv_https___arxiv_org_abs_2503_04857
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A kinetic-based regularization method for data science applications
Ganguly, Abhisek
Gabbana, Alessandro
Rao, Vybhav
Succi, Sauro
Ansumali, Santosh
Machine Learning
We propose a physics-based regularization technique for function learning, inspired by statistical mechanics. By drawing an analogy between optimizing the parameters of an interpolator and minimizing the energy of a system, we introduce corrections that impose constraints on the lower-order moments of the data distribution. This minimizes the discrepancy between the discrete and continuum representations of the data, in turn allowing to access more favorable energy landscapes, thus improving the accuracy of the interpolator. Our approach improves performance in both interpolation and regression tasks, even in high-dimensional spaces. Unlike traditional methods, it does not require empirical parameter tuning, making it particularly effective for handling noisy data. We also show that thanks to its local nature, the method offers computational and memory efficiency advantages over Radial Basis Function interpolators, especially for large datasets.
title A kinetic-based regularization method for data science applications
topic Machine Learning
url https://arxiv.org/abs/2503.04857