亲爱的研友该休息了!由于当前在线用户较少,发布求助请尽量完整地填写文献信息,科研通机器人24小时在线,伴您度过漫漫科研夜!身体可是革命的本钱,早点休息,好梦!

Zero-Shot Sparse Mixture of Low-Rank Experts Construction From Pre-Trained Foundation Models

可解释性 计算机科学 杠杆(统计) 机器学习 人工智能 可扩展性 子空间拓扑 适应性 深度学习 数据挖掘 干扰(通信) 数据建模 编码(集合论) 线性规划 线性模型 源代码 人工神经网络 混合模型 通过镜头测光 模式识别(心理学) 估计理论 上下文图像分类 传感器融合 预测建模 利用 高斯过程
作者
Anke Tang,Li Shen,Yong Luo,Shuai Xie,Han Hu,Lefei Zhang,Bo Du,Dacheng Tao
出处
期刊:IEEE Transactions on Pattern Analysis and Machine Intelligence [IEEE Computer Society]
卷期号:48 (2): 1145-1157 被引量:20
标识
DOI:10.1109/tpami.2025.3612480
摘要

Deep model training on extensive datasets is increasingly becoming cost-prohibitive, prompting the widespread adoption of deep model fusion techniques to leverage knowledge from pre-existing models. From simple weight averaging to more sophisticated methods like AdaMerging, model fusion effectively improves model performance and accelerates the development of new models. However, potential interference between parameters of individual models and the lack of interpretability in the fusion progress remain significant challenges. Existing methods often try to resolve the parameter interference issue by evaluating attributes of parameters, such as their magnitude or sign, or by parameter pruning. In this study, we begin by examining the fine-tuning of linear layers through the lens of subspace analysis and explicitly define parameter interference as an optimization problem to shed light on this subject. Subsequently, we introduce an innovative approach to model fusion called zero-shot Sparse MIxture of Low-rank Experts (SMILE) construction, which allows for the upscaling of source models into an MoE model without extra data or further training. Our approach relies on the observation that fine-tuning mostly keeps the important parts from the pre-training, but it uses less significant or unused areas to adapt to new tasks. Additionally, the issue of parameter interference, which is intrinsically challenging in the original parameter space, can be managed by expanding the dimensions. We conduct extensive experiments across diverse scenarios, such as image classification and text generation tasks, using full fine-tuning and LoRA fine-tuning, and we apply our method to large language models (CLIP models, Flan-T5 models, and Mistral-7B models), highlighting the adaptability and scalability of SMILE. For full fine-tuned models, about 50% additional parameters can achieve around 98-99% of the performance of eight individual fine-tuned ViT models, while for LoRA fine-tuned Flan-T5 models, maintaining 99% performance with only 2% extra parameters. Code is available athttps://github.com/tanganke/fusion_bench.
最长约 10秒,即可获得该文献文件

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
hhuajw完成签到,获得积分10
5秒前
8秒前
领导范儿的应助被zhengjia采纳,获得10
11秒前
SUNny发布了新的文献求助10
13秒前
Maestro_S的应助被爱学习的doge采纳,获得10
14秒前
彭于晏的应助被爱学习的doge采纳,获得10
14秒前
星辰大海的应助被爱学习的doge采纳,获得10
14秒前
15秒前
今后的应助被爱学习的doge采纳,获得10
15秒前
酷波er的应助被爱学习的doge采纳,获得10
15秒前
15秒前
上官若男的应助被爱学习的doge采纳,获得10
21秒前
星辰大海的应助被爱学习的doge采纳,获得10
21秒前
爆米花的应助被爱学习的doge采纳,获得10
21秒前
FashionBoy的应助被爱学习的doge采纳,获得10
21秒前
bkagyin的应助被爱学习的doge采纳,获得10
21秒前
传奇3的应助被爱学习的doge采纳,获得30
21秒前
小马甲的应助被爱学习的doge采纳,获得10
21秒前
斯文败类的应助被爱学习的doge采纳,获得10
22秒前
汉堡包的应助被爱学习的doge采纳,获得10
22秒前
22秒前
25秒前
乐乐的应助被爱学习的doge采纳,获得10
28秒前
丘比特的应助被爱学习的doge采纳,获得10
28秒前
bkagyin的应助被爱学习的doge采纳,获得10
28秒前
Orange的应助被爱学习的doge采纳,获得10
28秒前
领导范儿的应助被爱学习的doge采纳,获得30
28秒前
我是老大的应助被爱学习的doge采纳,获得10
28秒前
情怀的应助被爱学习的doge采纳,获得10
28秒前
打打的应助被爱学习的doge采纳,获得10
28秒前
香蕉觅云的应助被爱学习的doge采纳,获得10
28秒前
传奇3的应助被爱学习的doge采纳,获得10
28秒前
29秒前
31秒前
32秒前
无语的新之完成签到,获得积分10
32秒前
32秒前
32秒前
33秒前
33秒前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
拉伸高分子熔体中缠结行为的模拟和小角中子散射研究 1700
The World Atlas of Wine 9th edition 1000
奈特基础组织学:配套组织病理学(第4版)(Netter's Essential Histology With Correlated Histopathology) 800
John McDowells Bildungstheorie 800
Social Psychology in the Real World 600
八旗制度の研究 600
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 计算机科学 工程类 有机化学 化学工程 生物化学 物理 复合材料 内科学 催化作用 光电子学 人工智能 细胞生物学 无机化学 物理化学 数学
热门帖子
关注 科研通微信公众号,转发送积分 8179936
求助须知:如何正确求助?哪些是违规求助? 9570159
关于积分的说明 21808414
捐赠科研通 7775767
什么是DOI,文献DOI怎么找? 3433224
关于科研通互助平台的介绍 2557347
邀请新用户注册赠送积分活动 2478929