Human shape representations are not an emergent property of learning to classify objects.

IF 3.7 1区 心理学 Q1 PSYCHOLOGY, EXPERIMENTAL
Journal of Experimental Psychology: General Pub Date : 2023-12-01 Epub Date: 2023-09-11 DOI:10.1037/xge0001440
Gaurav Malhotra, Marin Dujmović, John Hummel, Jeffrey S Bowers
{"title":"Human shape representations are not an emergent property of learning to classify objects.","authors":"Gaurav Malhotra, Marin Dujmović, John Hummel, Jeffrey S Bowers","doi":"10.1037/xge0001440","DOIUrl":null,"url":null,"abstract":"<p><p>Humans are particularly sensitive to relationships between parts of objects. It remains unclear why this is. One hypothesis is that relational features are highly diagnostic of object categories and emerge as a result of learning to classify objects. We tested this by analyzing the internal representations of supervised convolutional neural networks (CNNs) trained to classify large sets of objects. We found that CNNs do not show the same sensitivity to relational changes as previously observed for human participants. Furthermore, when we precisely controlled the deformations to objects, human behavior was best predicted by the number of relational changes while CNNs were equally sensitive to all changes. Even changing the statistics of the learning environment by making relations uniquely diagnostic did not make networks more sensitive to relations in general. Our results show that learning to classify objects is not sufficient for the emergence of human shape representations. Instead, these results suggest that humans are selectively sensitive to relational changes because they build representations of distal objects from their retinal images and interpret relational changes as changes to these distal objects. This inferential process makes human shape representations qualitatively different from those of artificial neural networks optimized to perform image classification. (PsycInfo Database Record (c) 2023 APA, all rights reserved).</p>","PeriodicalId":15698,"journal":{"name":"Journal of Experimental Psychology: General","volume":" ","pages":"3380-3402"},"PeriodicalIF":3.7000,"publicationDate":"2023-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Experimental Psychology: General","FirstCategoryId":"102","ListUrlMain":"https://doi.org/10.1037/xge0001440","RegionNum":1,"RegionCategory":"心理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2023/9/11 0:00:00","PubModel":"Epub","JCR":"Q1","JCRName":"PSYCHOLOGY, EXPERIMENTAL","Score":null,"Total":0}
引用次数: 0

Abstract

Humans are particularly sensitive to relationships between parts of objects. It remains unclear why this is. One hypothesis is that relational features are highly diagnostic of object categories and emerge as a result of learning to classify objects. We tested this by analyzing the internal representations of supervised convolutional neural networks (CNNs) trained to classify large sets of objects. We found that CNNs do not show the same sensitivity to relational changes as previously observed for human participants. Furthermore, when we precisely controlled the deformations to objects, human behavior was best predicted by the number of relational changes while CNNs were equally sensitive to all changes. Even changing the statistics of the learning environment by making relations uniquely diagnostic did not make networks more sensitive to relations in general. Our results show that learning to classify objects is not sufficient for the emergence of human shape representations. Instead, these results suggest that humans are selectively sensitive to relational changes because they build representations of distal objects from their retinal images and interpret relational changes as changes to these distal objects. This inferential process makes human shape representations qualitatively different from those of artificial neural networks optimized to perform image classification. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

人类的形状表征并不是学习分类物体的一个新特性。
人类对物体各部分之间的关系特别敏感。目前还不清楚为什么会这样。一个假设是,关系特征是高度诊断对象类别,并出现作为学习分类对象的结果。我们通过分析监督卷积神经网络(cnn)的内部表示来测试这一点,cnn训练用于对大型对象集进行分类。我们发现cnn对关系变化的敏感度不如之前在人类参与者中观察到的那样高。此外,当我们精确控制物体的变形时,人类行为的最佳预测是通过关系变化的数量,而cnn对所有变化都同样敏感。即使通过使关系具有独特的诊断性来改变学习环境的统计数据,也不会使网络对一般关系更敏感。我们的研究结果表明,学习对物体进行分类是不足以产生人类形状表征的。相反,这些结果表明,人类对关系变化有选择性地敏感,因为他们从视网膜图像中构建远端物体的表征,并将关系变化解释为这些远端物体的变化。这种推理过程使得人类形状表征与用于图像分类的人工神经网络在本质上有所不同。(PsycInfo数据库记录(c) 2023 APA,版权所有)。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 求助全文
来源期刊
CiteScore
6.20
自引率
4.90%
发文量
300
期刊介绍: The Journal of Experimental Psychology: General publishes articles describing empirical work that bridges the traditional interests of two or more communities of psychology. The work may touch on issues dealt with in JEP: Learning, Memory, and Cognition, JEP: Human Perception and Performance, JEP: Animal Behavior Processes, or JEP: Applied, but may also concern issues in other subdisciplines of psychology, including social processes, developmental processes, psychopathology, neuroscience, or computational modeling. Articles in JEP: General may be longer than the usual journal publication if necessary, but shorter articles that bridge subdisciplines will also be considered.
×
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