Deep Stable Learning for Out-Of-Distribution Generalization

一般化 MNIST数据库 计算机科学 人工智能 判别式 虚假关系 深度学习 深层神经网络 机器学习 训练集 人工神经网络 分布(数学) 模式识别(心理学) 数学 数学分析
作者
Xingxuan Zhang,Peng Cui,Renzhe Xu,Linjun Zhou,Yue He,Zheyan Shen
标识
DOI:10.1109/cvpr46437.2021.00533
摘要

Approaches based on deep neural networks have achieved striking performance when testing data and training data share similar distribution, but can significantly fail otherwise. Therefore, eliminating the impact of distribution shifts between training and testing data is crucial for building performance-promising deep models. Conventional methods assume either the known heterogeneity of training data (e.g. domain labels) or the approximately equal capacities of different domains. In this paper, we consider a more challenging case where neither of the above assumptions holds. We propose to address this problem by removing the dependencies between features via learning weights for training samples, which helps deep models get rid of spurious correlations and, in turn, concentrate more on the true connection between discriminative features and labels. Extensive experiments clearly demonstrate the effectiveness of our method on multiple distribution generalization benchmarks compared with state-of-the-art counterparts. Through extensive experiments on distribution generalization benchmarks including PACS, VLCS, MNIST-M, and NICO, we show the effectiveness of our method compared with state-of-the-art counterparts.
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
Pendragon发布了新的文献求助10
刚刚
1秒前
2秒前
wh发布了新的文献求助10
3秒前
一朵梅花完成签到,获得积分10
4秒前
4秒前
春二虫完成签到,获得积分10
5秒前
斯文败类应助Pendragon采纳,获得10
6秒前
6秒前
flower发布了新的文献求助10
7秒前
8秒前
8秒前
八宝粥完成签到,获得积分10
9秒前
华仔应助21采纳,获得30
9秒前
沉默问夏完成签到 ,获得积分10
9秒前
9秒前
10秒前
羽翼发布了新的文献求助10
10秒前
11秒前
汉堡包应助Mydddg采纳,获得10
11秒前
科研小王发布了新的文献求助30
12秒前
12秒前
搜集达人应助lixl0725采纳,获得10
12秒前
14秒前
Lynn完成签到,获得积分10
14秒前
景阳完成签到,获得积分10
14秒前
shilong.yang发布了新的文献求助10
14秒前
英姑应助自由的白安采纳,获得10
14秒前
15秒前
科研通AI5应助妮妮采纳,获得10
16秒前
wh完成签到,获得积分10
16秒前
siri发布了新的文献求助30
17秒前
我不得依较完成签到,获得积分10
17秒前
念念发布了新的文献求助10
17秒前
执着半雪发布了新的文献求助10
18秒前
madcatalysis发布了新的文献求助10
18秒前
18秒前
xiao芳完成签到,获得积分10
18秒前
19秒前
caoyulongchn完成签到,获得积分10
20秒前
高分求助中
Algorithmic Mathematics in Machine Learning 500
Handbook of Innovations in Political Psychology 400
Mapping the Stars: Celebrity, Metonymy, and the Networked Politics of Identity 400
Visceral obesity is associated with clinical and inflammatory features of asthma: A prospective cohort study 300
Getting Published in SSCI Journals: 200+ Questions and Answers for Absolute Beginners 300
Engineering the boosting of the magnetic Purcell factor with a composite structure based on nanodisk and ring resonators 240
HVAC 1 toolkit : a toolkit for primary HVAC system energy calculation 200
热门求助领域 (近24小时)
化学 材料科学 医学 生物 工程类 有机化学 物理 生物化学 纳米技术 计算机科学 化学工程 内科学 复合材料 物理化学 电极 遗传学 量子力学 基因 冶金 催化作用
热门帖子
关注 科研通微信公众号,转发送积分 3839571
求助须知:如何正确求助?哪些是违规求助? 3381905
关于积分的说明 10520504
捐赠科研通 3101362
什么是DOI,文献DOI怎么找? 1708032
邀请新用户注册赠送积分活动 822099
科研通“疑难数据库(出版商)”最低求助积分说明 773174