Deep neural networks are easily fooled: High confidence predictions for unrecognizable images

Deep neural networks (DNNs) have recently been achieving state-of-the-art performance on a variety of pattern-recognition tasks, most notably visual classification problems. Given that DNNs are now able to classify objects in images with near-human-level performance, questions naturally arise as to...

Full description

Saved in:
Bibliographic Details
Published in2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) pp. 427 - 436
Main Authors Anh Nguyen, Yosinski, Jason, Clune, Jeff
Format Conference Proceeding Journal Article
LanguageEnglish
Published IEEE 01.06.2015
Subjects
Online AccessGet full text
ISSN1063-6919
1063-6919
DOI10.1109/CVPR.2015.7298640

Cover

Abstract Deep neural networks (DNNs) have recently been achieving state-of-the-art performance on a variety of pattern-recognition tasks, most notably visual classification problems. Given that DNNs are now able to classify objects in images with near-human-level performance, questions naturally arise as to what differences remain between computer and human vision. A recent study [30] revealed that changing an image (e.g. of a lion) in a way imperceptible to humans can cause a DNN to label the image as something else entirely (e.g. mislabeling a lion a library). Here we show a related result: it is easy to produce images that are completely unrecognizable to humans, but that state-of-the-art DNNs believe to be recognizable objects with 99.99% confidence (e.g. labeling with certainty that white noise static is a lion). Specifically, we take convolutional neural networks trained to perform well on either the ImageNet or MNIST datasets and then find images with evolutionary algorithms or gradient ascent that DNNs label with high confidence as belonging to each dataset class. It is possible to produce images totally unrecognizable to human eyes that DNNs believe with near certainty are familiar objects, which we call "fooling images" (more generally, fooling examples). Our results shed light on interesting differences between human vision and current DNNs, and raise questions about the generality of DNN computer vision.
AbstractList Deep neural networks (DNNs) have recently been achieving state-of-the-art performance on a variety of pattern-recognition tasks, most notably visual classification problems. Given that DNNs are now able to classify objects in images with near-human-level performance, questions naturally arise as to what differences remain between computer and human vision. A recent study [30] revealed that changing an image (e.g. of a lion) in a way imperceptible to humans can cause a DNN to label the image as something else entirely (e.g. mislabeling a lion a library). Here we show a related result: it is easy to produce images that are completely unrecognizable to humans, but that state-of-the-art DNNs believe to be recognizable objects with 99.99% confidence (e.g. labeling with certainty that white noise static is a lion). Specifically, we take convolutional neural networks trained to perform well on either the ImageNet or MNIST datasets and then find images with evolutionary algorithms or gradient ascent that DNNs label with high confidence as belonging to each dataset class. It is possible to produce images totally unrecognizable to human eyes that DNNs believe with near certainty are familiar objects, which we call "fooling images" (more generally, fooling examples). Our results shed light on interesting differences between human vision and current DNNs, and raise questions about the generality of DNN computer vision.
Author Anh Nguyen
Yosinski, Jason
Clune, Jeff
Author_xml – sequence: 1
  surname: Anh Nguyen
  fullname: Anh Nguyen
  email: anguyen8@uwyo.edu
  organization: Univ. of Wyoming, Laramie, WY, USA
– sequence: 2
  givenname: Jason
  surname: Yosinski
  fullname: Yosinski, Jason
  email: yosinski@cs.cornell.edu
  organization: Cornell Univ., Ithaca, NY, USA
– sequence: 3
  givenname: Jeff
  surname: Clune
  fullname: Clune, Jeff
  email: jeffclune@uwyo.edu
  organization: Univ. of Wyoming, Laramie, WY, USA
BookMark eNqN0E1LAzEQBuAoFWxrf4B4ydHL1mSzmzTepH5UKCii3mTJx2wNbpM12UXqr3ehHjx6emfgYZiZCRr54AGhU0rmlBJ5sXx9fJrnhJZzkcsFL8gBmtCCC8bl0ByiMSWcZVxSOfpTH6NZSk4TRshCypyM0ds1QIs99FE1Q3RfIX4krCJgUMk1O1yH0IC9xCu3eccm-NpZ8AZwG8E607ng02Ai7n0EEzbefSvdAHZbtYF0go5q1SSY_eYUvdzePC9X2frh7n55tc4cI6LLuNayyCFnwLWtS2Y4KY3MqTBW5YRTbYfdmbbcalUYoVhtgdGBS1uqouBsis73c9sYPntIXbV1yUDTKA-hTxUVgrCClZz-g_JFKXheioGe7akDgKqNw01xV_3-m_0AQUl0jQ
ContentType Conference Proceeding
Journal Article
DBID 6IE
6IH
CBEJK
RIE
RIO
7QO
8FD
FR3
P64
7SC
JQ2
L7M
L~C
L~D
DOI 10.1109/CVPR.2015.7298640
DatabaseName IEEE Electronic Library (IEL) Conference Proceedings
IEEE Proceedings Order Plan (POP) 1998-present by volume
IEEE Xplore All Conference Proceedings
Consulter via IEEE Xplore
IEEE Proceedings Order Plans (POP) 1998-present
Biotechnology Research Abstracts
Technology Research Database
Engineering Research Database
Biotechnology and BioEngineering Abstracts
Computer and Information Systems Abstracts
ProQuest Computer Science Collection
Advanced Technologies Database with Aerospace
Computer and Information Systems Abstracts – Academic
Computer and Information Systems Abstracts Professional
DatabaseTitle Engineering Research Database
Biotechnology Research Abstracts
Technology Research Database
Biotechnology and BioEngineering Abstracts
Computer and Information Systems Abstracts
Computer and Information Systems Abstracts – Academic
Advanced Technologies Database with Aerospace
ProQuest Computer Science Collection
Computer and Information Systems Abstracts Professional
DatabaseTitleList Computer and Information Systems Abstracts

Engineering Research Database
Database_xml – sequence: 1
  dbid: RIE
  name: IEEE Xplore
  url: https://proxy.k.utb.cz/login?url=https://ieeexplore.ieee.org/
  sourceTypes: Publisher
DeliveryMethod fulltext_linktorsrc
Discipline Applied Sciences
Computer Science
EISBN 1467369640
9781467369640
EISSN 1063-6919
EndPage 436
ExternalDocumentID 7298640
Genre orig-research
GroupedDBID 23M
29F
29O
6IE
6IH
6IK
ABDPE
ACGFS
ALMA_UNASSIGNED_HOLDINGS
CBEJK
IPLJI
M43
RIE
RIO
RNS
7QO
8FD
FR3
P64
7SC
JQ2
L7M
L~C
L~D
ID FETCH-LOGICAL-i307t-6bb942e23e6bdf53c605c9217cda2061bd9193bd6dba4c7a3fde3123e9d5a4463
IEDL.DBID RIE
ISSN 1063-6919
IngestDate Tue Aug 05 09:33:32 EDT 2025
Fri Sep 05 11:44:21 EDT 2025
Wed Aug 27 02:49:08 EDT 2025
IsPeerReviewed false
IsScholarly true
Language English
LinkModel DirectLink
MergedId FETCHMERGED-LOGICAL-i307t-6bb942e23e6bdf53c605c9217cda2061bd9193bd6dba4c7a3fde3123e9d5a4463
Notes ObjectType-Article-2
SourceType-Scholarly Journals-1
ObjectType-Conference-1
ObjectType-Feature-3
content type line 23
SourceType-Conference Papers & Proceedings-2
PQID 1768576257
PQPubID 23462
PageCount 10
ParticipantIDs ieee_primary_7298640
proquest_miscellaneous_1768576257
proquest_miscellaneous_1770343561
PublicationCentury 2000
PublicationDate 20150601
PublicationDateYYYYMMDD 2015-06-01
PublicationDate_xml – month: 06
  year: 2015
  text: 20150601
  day: 01
PublicationDecade 2010
PublicationTitle 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
PublicationTitleAbbrev CVPR
PublicationYear 2015
Publisher IEEE
Publisher_xml – name: IEEE
SSID ssib030089920
ssj0023720
ssj0003211698
Score 2.5710287
Snippet Deep neural networks (DNNs) have recently been achieving state-of-the-art performance on a variety of pattern-recognition tasks, most notably visual...
SourceID proquest
ieee
SourceType Aggregation Database
Publisher
StartPage 427
SubjectTerms Biomedical imaging
Classification
Computer vision
Confidence
Human
Keyboards
Labels
Neural networks
State of the art
Volcanoes
Title Deep neural networks are easily fooled: High confidence predictions for unrecognizable images
URI https://ieeexplore.ieee.org/document/7298640
https://www.proquest.com/docview/1768576257
https://www.proquest.com/docview/1770343561
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
link http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwjV1LS8NAEF6qJ0--sb5YwaOpbTbZdr1WiwiKiJVepOxjAsWalKY52F_vTLKpoCKeEpYlj83s7Hz5vtlh7NzRplk9lQQaHKXkgA1M3NVBz0ZdbTDCh1JEc_8gb4fR3SgeNdjFKhcGAErxGbTotOTyXWYL-lV2iYFgT0YI0NfQzKpcrdp2RJv4Kx_6kBcWiGykWjEKIVVjKZlPKQKpOsoznJ22uuy_PD6RyCtu-Rv4Sis_3HO55gw22X39tJXU5K1VLEzLLr9t5Pjf19lie1_ZffxxtW5tswakO2zTh6PcT_Ycm-qKD3XbLnu9Bphx2gNTT_FQKshzrufAQeeT6QdPsmwK7oqTfIQj1E6qmqV8NidCqLRx7DPnReqVS0tK3eKTd3Rr-R4bDm6e-7eBL9AQTNA1LAJpjIpCCAVI45JYWMRGViHIsU6HGCgYhwMtDNWs0pHtapE4ELhUgnKxRhwq9tl6mqVwwHjPCiGFSyx0k8jEsdHEs8U2TLQyNoyabJcGcDyr9uAY-7FrsrP6E41xXhDZoVPIinzcQRyFWAo90l990N9hvCg7h79f_ohtkF1U0rBjtr6YF3CCQcjCnJbW9wkBLNmd
linkProvider IEEE
linkToHtml http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwjV1LT9wwELYQPcAJWqgKBWokjmRh49gbc-WhbWERQlBxqSI_JtKKbbLabA7dX9-ZxFkkqFBPiSwrD8cez5fvmxnGjjwlzUp1HhnwFJIDLrJyYKLUJQNj0cOHRkQzulXDx-THk3xaYcfLWBgAaMRn0KPThsv3pavpV9kJOoKpShCgf5CIKtI2WqubPeKUGKzg_JAdFohtlF5yCjHVY2m4TyUipfs6cJz9U31y_vPunmReshduEWqtvDHQza5ztcFG3fO2YpPnXj23Pbd4lcrxf19ok22_xPfxu-XO9ZGtQPGJbQSHlIflXmFTV_Oha9tivy4AppyyYJoJHhoNecXNDDiYajz5w_OynIA_4yQg4Qi287ZqKZ_OiBJqZjn2mfG6CNqlBQVv8fFvNGzVNnu8unw4H0ahREM0RuMwj5S1OokhFqCsz6VwiI6cRpjjvInRVbAeB1pYqlplEjcwIvcgcLME7aVBJCo-s9WiLOAL46kTQgmfOxjkiZXSGmLapItzo62Lkx22RQOYTdssHFkYux122H2iDFcG0R2mgLKusj4iKURTaJPe64MWDz1G1d_99-W_sbXhw-gmu_l-e_2VrdMcaYVie2x1PqthH12SuT1oZuJfnXzc8A
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=proceeding&rft.title=2015+IEEE+Conference+on+Computer+Vision+and+Pattern+Recognition+%28CVPR%29&rft.atitle=Deep+neural+networks+are+easily+fooled%3A+High+confidence+predictions+for+unrecognizable+images&rft.au=Anh+Nguyen&rft.au=Yosinski%2C+Jason&rft.au=Clune%2C+Jeff&rft.date=2015-06-01&rft.pub=IEEE&rft.issn=1063-6919&rft.eissn=1063-6919&rft.spage=427&rft.epage=436&rft_id=info:doi/10.1109%2FCVPR.2015.7298640&rft.externalDocID=7298640
thumbnail_l http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=1063-6919&client=summon
thumbnail_m http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=1063-6919&client=summon
thumbnail_s http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=1063-6919&client=summon