Distilling vision-language pre-training models with modality-specific meta-learning

IF 7.2 1区 计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE
Xinge Ma, Jin Wang, Xuejie Zhang
{"title":"Distilling vision-language pre-training models with modality-specific meta-learning","authors":"Xinge Ma,&nbsp;Jin Wang,&nbsp;Xuejie Zhang","doi":"10.1016/j.knosys.2025.113300","DOIUrl":null,"url":null,"abstract":"<div><div>Vision-language pre-training (VLP) models have exhibited excellent performance on diverse vision-language tasks, while the ensuing large-scale model parameters greatly limit their application. Knowledge distillation (KD) makes it possible to apply a VLP model to scenarios with limited resources and real-time responses by transferring the knowledge from it (<em>i.e.</em>, teacher) into a lightweight one (<em>i.e.</em>, student). However, existing KD methods are primarily designed for unimodal models and thus fail to realize their full potential when migrating them to distill VLP models considering the presence of multiple modalities. Moreover, these KD strategies only unilaterally force the student model to approach the output feature maps generated by the teacher model while ignoring the deeper correlations between them, which may hinder effective knowledge transfer. To tackle these issues, we propose MMKD, a multimodal <strong>K</strong>nowledge <strong>D</strong>istillation method with <strong>M</strong>odality-specific <strong>M</strong>eta-learning, in which the training objective of the teacher model is converted to optimize the teaching ability for knowledge transfer through feedback from the student model. Meanwhile, to disentangle mutual interference between different modalities when applying KD, the modality-specific distillation objective is designed to encourage the student model to learn the teacher’s knowledge from different modalities. By progressively optimizing the teacher model towards the direction of maximizing the student’s performance, more appropriate soft labels are generated to help the student model learn across different modalities, leading to improved performance. Experiments on three types of VLP models across different downstream tasks demonstrate that the superiority of the proposed method in compressing VLP models.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"315 ","pages":"Article 113300"},"PeriodicalIF":7.2000,"publicationDate":"2025-03-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Knowledge-Based Systems","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0950705125003478","RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE","Score":null,"Total":0}
引用次数: 0

Abstract

Vision-language pre-training (VLP) models have exhibited excellent performance on diverse vision-language tasks, while the ensuing large-scale model parameters greatly limit their application. Knowledge distillation (KD) makes it possible to apply a VLP model to scenarios with limited resources and real-time responses by transferring the knowledge from it (i.e., teacher) into a lightweight one (i.e., student). However, existing KD methods are primarily designed for unimodal models and thus fail to realize their full potential when migrating them to distill VLP models considering the presence of multiple modalities. Moreover, these KD strategies only unilaterally force the student model to approach the output feature maps generated by the teacher model while ignoring the deeper correlations between them, which may hinder effective knowledge transfer. To tackle these issues, we propose MMKD, a multimodal Knowledge Distillation method with Modality-specific Meta-learning, in which the training objective of the teacher model is converted to optimize the teaching ability for knowledge transfer through feedback from the student model. Meanwhile, to disentangle mutual interference between different modalities when applying KD, the modality-specific distillation objective is designed to encourage the student model to learn the teacher’s knowledge from different modalities. By progressively optimizing the teacher model towards the direction of maximizing the student’s performance, more appropriate soft labels are generated to help the student model learn across different modalities, leading to improved performance. Experiments on three types of VLP models across different downstream tasks demonstrate that the superiority of the proposed method in compressing VLP models.
求助全文
约1分钟内获得全文 求助全文
来源期刊
Knowledge-Based Systems
Knowledge-Based Systems 工程技术-计算机:人工智能
CiteScore
14.80
自引率
12.50%
发文量
1245
审稿时长
7.8 months
期刊介绍: Knowledge-Based Systems, an international and interdisciplinary journal in artificial intelligence, publishes original, innovative, and creative research results in the field. It focuses on knowledge-based and other artificial intelligence techniques-based systems. The journal aims to support human prediction and decision-making through data science and computation techniques, provide a balanced coverage of theory and practical study, and encourage the development and implementation of knowledge-based intelligence models, methods, systems, and software tools. Applications in business, government, education, engineering, and healthcare are emphasized.
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信