Erasing the Bias: Fine-Tuning Foundation Models for Semi-Supervised Learning

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Gan, Kai, Wei, Tong
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911881995747328
author Gan, Kai
Wei, Tong
author_facet Gan, Kai
Wei, Tong
contents Semi-supervised learning (SSL) has witnessed remarkable progress, resulting in the emergence of numerous method variations. However, practitioners often encounter challenges when attempting to deploy these methods due to their subpar performance. In this paper, we present a novel SSL approach named FineSSL that significantly addresses this limitation by adapting pre-trained foundation models. We identify the aggregated biases and cognitive deviation problems inherent in foundation models, and propose a simple yet effective solution by imposing balanced margin softmax and decoupled label smoothing. Through extensive experiments, we demonstrate that FineSSL sets a new state of the art for SSL on multiple benchmark datasets, reduces the training cost by over six times, and can seamlessly integrate various fine-tuning and modern SSL algorithms. The source code is available at https://github.com/Gank0078/FineSSL.
format Preprint
id arxiv_https___arxiv_org_abs_2405_11756
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Erasing the Bias: Fine-Tuning Foundation Models for Semi-Supervised Learning
Gan, Kai
Wei, Tong
Machine Learning
Semi-supervised learning (SSL) has witnessed remarkable progress, resulting in the emergence of numerous method variations. However, practitioners often encounter challenges when attempting to deploy these methods due to their subpar performance. In this paper, we present a novel SSL approach named FineSSL that significantly addresses this limitation by adapting pre-trained foundation models. We identify the aggregated biases and cognitive deviation problems inherent in foundation models, and propose a simple yet effective solution by imposing balanced margin softmax and decoupled label smoothing. Through extensive experiments, we demonstrate that FineSSL sets a new state of the art for SSL on multiple benchmark datasets, reduces the training cost by over six times, and can seamlessly integrate various fine-tuning and modern SSL algorithms. The source code is available at https://github.com/Gank0078/FineSSL.
title Erasing the Bias: Fine-Tuning Foundation Models for Semi-Supervised Learning
topic Machine Learning
url https://arxiv.org/abs/2405.11756