LPT++:长尾专家混合物的高效训练

Bowen Dong, Pan Zhou, Wangmeng Zuo
{"title":"LPT++:长尾专家混合物的高效训练","authors":"Bowen Dong, Pan Zhou, Wangmeng Zuo","doi":"arxiv-2409.11323","DOIUrl":null,"url":null,"abstract":"We introduce LPT++, a comprehensive framework for long-tailed classification\nthat combines parameter-efficient fine-tuning (PEFT) with a learnable model\nensemble. LPT++ enhances frozen Vision Transformers (ViTs) through the\nintegration of three core components. The first is a universal long-tailed\nadaptation module, which aggregates long-tailed prompts and visual adapters to\nadapt the pretrained model to the target domain, meanwhile improving its\ndiscriminative ability. The second is the mixture of long-tailed experts\nframework with a mixture-of-experts (MoE) scorer, which adaptively calculates\nreweighting coefficients for confidence scores from both visual-only and\nvisual-language (VL) model experts to generate more accurate predictions.\nFinally, LPT++ employs a three-phase training framework, wherein each critical\nmodule is learned separately, resulting in a stable and effective long-tailed\nclassification training paradigm. Besides, we also propose the simple version\nof LPT++ namely LPT, which only integrates visual-only pretrained ViT and\nlong-tailed prompts to formulate a single model method. LPT can clearly\nillustrate how long-tailed prompts works meanwhile achieving comparable\nperformance without VL pretrained models. Experiments show that, with only ~1%\nextra trainable parameters, LPT++ achieves comparable accuracy against all the\ncounterparts.","PeriodicalId":501130,"journal":{"name":"arXiv - CS - Computer Vision and Pattern Recognition","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2024-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"LPT++: Efficient Training on Mixture of Long-tailed Experts\",\"authors\":\"Bowen Dong, Pan Zhou, Wangmeng Zuo\",\"doi\":\"arxiv-2409.11323\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"We introduce LPT++, a comprehensive framework for long-tailed classification\\nthat combines parameter-efficient fine-tuning (PEFT) with a learnable model\\nensemble. LPT++ enhances frozen Vision Transformers (ViTs) through the\\nintegration of three core components. The first is a universal long-tailed\\nadaptation module, which aggregates long-tailed prompts and visual adapters to\\nadapt the pretrained model to the target domain, meanwhile improving its\\ndiscriminative ability. The second is the mixture of long-tailed experts\\nframework with a mixture-of-experts (MoE) scorer, which adaptively calculates\\nreweighting coefficients for confidence scores from both visual-only and\\nvisual-language (VL) model experts to generate more accurate predictions.\\nFinally, LPT++ employs a three-phase training framework, wherein each critical\\nmodule is learned separately, resulting in a stable and effective long-tailed\\nclassification training paradigm. Besides, we also propose the simple version\\nof LPT++ namely LPT, which only integrates visual-only pretrained ViT and\\nlong-tailed prompts to formulate a single model method. LPT can clearly\\nillustrate how long-tailed prompts works meanwhile achieving comparable\\nperformance without VL pretrained models. Experiments show that, with only ~1%\\nextra trainable parameters, LPT++ achieves comparable accuracy against all the\\ncounterparts.\",\"PeriodicalId\":501130,\"journal\":{\"name\":\"arXiv - CS - Computer Vision and Pattern Recognition\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-09-17\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - CS - Computer Vision and Pattern Recognition\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2409.11323\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - CS - Computer Vision and Pattern Recognition","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2409.11323","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0

摘要

我们介绍了 LPT++,这是一种用于长尾分类的综合框架,它将参数高效微调(PEFT)与可学习的模型组合结合在一起。LPT++ 通过整合三个核心组件,增强了冷冻视觉转换器(ViTs)。第一个是通用长尾适应模块,它将长尾提示和视觉适配器聚合在一起,使预训练模型适应目标领域,同时提高其识别能力。其次是长尾专家混合框架(mixed of long-tailed expertsframework)和专家混合评分器(mixed-of-experts,MoE),该评分器可以自适应地计算来自纯视觉和视觉语言(VL)模型专家的置信度评分的加权系数,从而生成更准确的预测。最后,LPT++ 采用了三阶段训练框架,其中每个关键模块都是单独学习的,从而形成了稳定有效的长尾分类训练范式。此外,我们还提出了LPT++的简易版本,即LPT,它只集成了纯视觉预训练的ViT和长尾提示,形成了单一的模型方法。LPT 可以清楚地展示长尾提示是如何工作的,同时在没有 VL 预训练模型的情况下也能取得相当的性能。实验表明,只需增加 ~1% 的可训练参数,LPT++ 就能达到与所有同行相当的准确率。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
LPT++: Efficient Training on Mixture of Long-tailed Experts
We introduce LPT++, a comprehensive framework for long-tailed classification that combines parameter-efficient fine-tuning (PEFT) with a learnable model ensemble. LPT++ enhances frozen Vision Transformers (ViTs) through the integration of three core components. The first is a universal long-tailed adaptation module, which aggregates long-tailed prompts and visual adapters to adapt the pretrained model to the target domain, meanwhile improving its discriminative ability. The second is the mixture of long-tailed experts framework with a mixture-of-experts (MoE) scorer, which adaptively calculates reweighting coefficients for confidence scores from both visual-only and visual-language (VL) model experts to generate more accurate predictions. Finally, LPT++ employs a three-phase training framework, wherein each critical module is learned separately, resulting in a stable and effective long-tailed classification training paradigm. Besides, we also propose the simple version of LPT++ namely LPT, which only integrates visual-only pretrained ViT and long-tailed prompts to formulate a single model method. LPT can clearly illustrate how long-tailed prompts works meanwhile achieving comparable performance without VL pretrained models. Experiments show that, with only ~1% extra trainable parameters, LPT++ achieves comparable accuracy against all the counterparts.
求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
自引率
0.00%
发文量
0
×
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