Modulate Your Spectrum in Self-Supervised Learning

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Weng, Xi, Ni, Yunhao, Song, Tengwei, Luo, Jie, Anwer, Rao Muhammad, Khan, Salman, Khan, Fahad Shahbaz, Huang, Lei
Format: Preprint
Veröffentlicht: 2023
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911761399021568
author Weng, Xi
Ni, Yunhao
Song, Tengwei
Luo, Jie
Anwer, Rao Muhammad
Khan, Salman
Khan, Fahad Shahbaz
Huang, Lei
author_facet Weng, Xi
Ni, Yunhao
Song, Tengwei
Luo, Jie
Anwer, Rao Muhammad
Khan, Salman
Khan, Fahad Shahbaz
Huang, Lei
contents Whitening loss offers a theoretical guarantee against feature collapse in self-supervised learning (SSL) with joint embedding architectures. Typically, it involves a hard whitening approach, transforming the embedding and applying loss to the whitened output. In this work, we introduce Spectral Transformation (ST), a framework to modulate the spectrum of embedding and to seek for functions beyond whitening that can avoid dimensional collapse. We show that whitening is a special instance of ST by definition, and our empirical investigations unveil other ST instances capable of preventing collapse. Additionally, we propose a novel ST instance named IterNorm with trace loss (INTL). Theoretical analysis confirms INTL's efficacy in preventing collapse and modulating the spectrum of embedding toward equal-eigenvalues during optimization. Our experiments on ImageNet classification and COCO object detection demonstrate INTL's potential in learning superior representations. The code is available at https://github.com/winci-ai/INTL.
format Preprint
id arxiv_https___arxiv_org_abs_2305_16789
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Modulate Your Spectrum in Self-Supervised Learning
Weng, Xi
Ni, Yunhao
Song, Tengwei
Luo, Jie
Anwer, Rao Muhammad
Khan, Salman
Khan, Fahad Shahbaz
Huang, Lei
Machine Learning
Computer Vision and Pattern Recognition
Signal Processing
Whitening loss offers a theoretical guarantee against feature collapse in self-supervised learning (SSL) with joint embedding architectures. Typically, it involves a hard whitening approach, transforming the embedding and applying loss to the whitened output. In this work, we introduce Spectral Transformation (ST), a framework to modulate the spectrum of embedding and to seek for functions beyond whitening that can avoid dimensional collapse. We show that whitening is a special instance of ST by definition, and our empirical investigations unveil other ST instances capable of preventing collapse. Additionally, we propose a novel ST instance named IterNorm with trace loss (INTL). Theoretical analysis confirms INTL's efficacy in preventing collapse and modulating the spectrum of embedding toward equal-eigenvalues during optimization. Our experiments on ImageNet classification and COCO object detection demonstrate INTL's potential in learning superior representations. The code is available at https://github.com/winci-ai/INTL.
title Modulate Your Spectrum in Self-Supervised Learning
topic Machine Learning
Computer Vision and Pattern Recognition
Signal Processing
url https://arxiv.org/abs/2305.16789