Variance-Covariance Regularization Improves Representation Learning

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Zhu, Jiachen, Evtimova, Katrina, Chen, Yubei, Shwartz-Ziv, Ravid, LeCun, Yann
Natura: Preprint
Pubblicazione: 2023
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916134706479104
author Zhu, Jiachen
Evtimova, Katrina
Chen, Yubei
Shwartz-Ziv, Ravid
LeCun, Yann
author_facet Zhu, Jiachen
Evtimova, Katrina
Chen, Yubei
Shwartz-Ziv, Ravid
LeCun, Yann
contents Transfer learning plays a key role in advancing machine learning models, yet conventional supervised pretraining often undermines feature transferability by prioritizing features that minimize the pretraining loss. In this work, we adapt a self-supervised learning regularization technique from the VICReg method to supervised learning contexts, introducing Variance-Covariance Regularization (VCReg). This adaptation encourages the network to learn high-variance, low-covariance representations, promoting learning more diverse features. We outline best practices for an efficient implementation of our framework, including applying it to the intermediate representations. Through extensive empirical evaluation, we demonstrate that our method significantly enhances transfer learning for images and videos, achieving state-of-the-art performance across numerous tasks and datasets. VCReg also improves performance in scenarios like long-tail learning and hierarchical classification. Additionally, we show its effectiveness may stem from its success in addressing challenges like gradient starvation and neural collapse. In summary, VCReg offers a universally applicable regularization framework that significantly advances transfer learning and highlights the connection between gradient starvation, neural collapse, and feature transferability.
format Preprint
id arxiv_https___arxiv_org_abs_2306_13292
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Variance-Covariance Regularization Improves Representation Learning
Zhu, Jiachen
Evtimova, Katrina
Chen, Yubei
Shwartz-Ziv, Ravid
LeCun, Yann
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Transfer learning plays a key role in advancing machine learning models, yet conventional supervised pretraining often undermines feature transferability by prioritizing features that minimize the pretraining loss. In this work, we adapt a self-supervised learning regularization technique from the VICReg method to supervised learning contexts, introducing Variance-Covariance Regularization (VCReg). This adaptation encourages the network to learn high-variance, low-covariance representations, promoting learning more diverse features. We outline best practices for an efficient implementation of our framework, including applying it to the intermediate representations. Through extensive empirical evaluation, we demonstrate that our method significantly enhances transfer learning for images and videos, achieving state-of-the-art performance across numerous tasks and datasets. VCReg also improves performance in scenarios like long-tail learning and hierarchical classification. Additionally, we show its effectiveness may stem from its success in addressing challenges like gradient starvation and neural collapse. In summary, VCReg offers a universally applicable regularization framework that significantly advances transfer learning and highlights the connection between gradient starvation, neural collapse, and feature transferability.
title Variance-Covariance Regularization Improves Representation Learning
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2306.13292