Exploring Representation Invariance in Finetuning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zu, Wenqiang, Xie, Shenghao, Chen, Hao, Chen, Zhiqiang, Hu, Liwen, Xi, Yuanhao, Liang, Yiming, Ye, Junliang, Lei, Bo, Huang, Tiejun, Li, Guoqi, Ma, Lei
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914073988300800
author Zu, Wenqiang
Xie, Shenghao
Chen, Hao
Chen, Zhiqiang
Hu, Liwen
Xi, Yuanhao
Liang, Yiming
Ye, Junliang
Lei, Bo
Huang, Tiejun
Li, Guoqi
Ma, Lei
author_facet Zu, Wenqiang
Xie, Shenghao
Chen, Hao
Chen, Zhiqiang
Hu, Liwen
Xi, Yuanhao
Liang, Yiming
Ye, Junliang
Lei, Bo
Huang, Tiejun
Li, Guoqi
Ma, Lei
contents Foundation models pretrained on large-scale natural images are widely adapted to various cross-domain low-resource downstream tasks, benefiting from generalizable and transferable patterns captured by their representations. However, these representations are later found to gradually vanish during finetuning, accompanied by a degradation of model's original generalizability. In this paper, we argue that such tasks can be effectively adapted without sacrificing the benefits of pretrained representations. We approach this by introducing \textit{Representation Invariance FineTuning (RIFT)}, a regularization that maximizes the representation similarity between pretrained and finetuned models by leveraging orthogonal invariance of manifolds in a computationally efficient way. Experiments demonstrate that our method is compatible with mainstream finetuning methods, offering competitive or even enhanced performance and better preservation of the generalizability.
format Preprint
id arxiv_https___arxiv_org_abs_2503_07399
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Exploring Representation Invariance in Finetuning
Zu, Wenqiang
Xie, Shenghao
Chen, Hao
Chen, Zhiqiang
Hu, Liwen
Xi, Yuanhao
Liang, Yiming
Ye, Junliang
Lei, Bo
Huang, Tiejun
Li, Guoqi
Ma, Lei
Computer Vision and Pattern Recognition
Foundation models pretrained on large-scale natural images are widely adapted to various cross-domain low-resource downstream tasks, benefiting from generalizable and transferable patterns captured by their representations. However, these representations are later found to gradually vanish during finetuning, accompanied by a degradation of model's original generalizability. In this paper, we argue that such tasks can be effectively adapted without sacrificing the benefits of pretrained representations. We approach this by introducing \textit{Representation Invariance FineTuning (RIFT)}, a regularization that maximizes the representation similarity between pretrained and finetuned models by leveraging orthogonal invariance of manifolds in a computationally efficient way. Experiments demonstrate that our method is compatible with mainstream finetuning methods, offering competitive or even enhanced performance and better preservation of the generalizability.
title Exploring Representation Invariance in Finetuning
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2503.07399