Robust deep learning object recognition models rely on low frequency information in natural images

Simple item page
dc.citation.articleNumbere1010932en_US
dc.citation.issueNumber3en_US
dc.citation.journalTitlePLOS Computational Biologyen_US
dc.citation.volumeNumber19en_US
dc.contributor.authorLi, Zheen_US
dc.contributor.authorCaro, Josue Ortegaen_US
dc.contributor.authorRusak, Evgeniaen_US
dc.contributor.authorBrendel, Wielanden_US
dc.contributor.authorBethge, Matthiasen_US
dc.contributor.authorAnselmi, Fabioen_US
dc.contributor.authorPatel, Ankit B.en_US
dc.contributor.authorTolias, Andreas S.en_US
dc.contributor.authorPitkow, Xaqen_US
dc.date.accessioned2023-04-25T14:48:12Zen_US
dc.date.available2023-04-25T14:48:12Zen_US
dc.date.issued2023en_US
dc.description.abstractMachine learning models have difficulty generalizing to data outside of the distribution they were trained on. In particular, vision models are usually vulnerable to adversarial attacks or common corruptions, to which the human visual system is robust. Recent studies have found that regularizing machine learning models to favor brain-like representations can improve model robustness, but it is unclear why. We hypothesize that the increased model robustness is partly due to the low spatial frequency preference inherited from the neural representation. We tested this simple hypothesis with several frequency-oriented analyses, including the design and use of hybrid images to probe model frequency sensitivity directly. We also examined many other publicly available robust models that were trained on adversarial images or with data augmentation, and found that all these robust models showed a greater preference to low spatial frequency information. We show that preprocessing by blurring can serve as a defense mechanism against both adversarial attacks and common corruptions, further confirming our hypothesis and demonstrating the utility of low spatial frequency information in robust object recognition.en_US
dc.identifier.citationLi, Zhe, Caro, Josue Ortega, Rusak, Evgenia, et al.. "Robust deep learning object recognition models rely on low frequency information in natural images." <i>PLOS Computational Biology,</i> 19, no. 3 (2023) PLOS: https://doi.org/10.1371/journal.pcbi.1010932.en_US
dc.identifier.digitaljournal-pcbi-1010932en_US
dc.identifier.doihttps://doi.org/10.1371/journal.pcbi.1010932en_US
dc.identifier.urihttps://hdl.handle.net/1911/114842en_US
dc.language.isoengen_US
dc.publisherPLOSen_US
dc.rightsThis is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.en_US
dc.rights.urihttps://creativecommons.org/licenses/by/4.0/en_US
dc.titleRobust deep learning object recognition models rely on low frequency information in natural imagesen_US
dc.typeJournal articleen_US
dc.type.dcmiTexten_US
dc.type.publicationpublisher versionen_US
Files
Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
journal-pcbi-1010932.pdf
Size:
2.02 MB
Format:
Adobe Portable Document Format
Download
Collections
Faculty Publications
ECE Publications