Shared texture-like representations underlie deep neural network alignment with human visual processing.

Λεπτομέρειες βιβλιογραφικής εγγραφής
Τίτλος:	Shared texture-like representations underlie deep neural network alignment with human visual processing.
Συγγραφείς:	Loke J; Department of Psychology, University of Amsterdam, Nieuwe Achtergracht 129-B, 1018WS Amsterdam, the Netherlands; Amsterdam Brain & Cognition (ABC) Center, University of Amsterdam, Nieuwe Achtergracht 129-B, 1018WS Amsterdam, the Netherlands. Electronic address: j.loke@uva.nl., Sörensen LKA; Department of Brain and Cognitive Sciences, McGovern Institute for Brain Research, Massachusetts Institute of Technology, Building 46, 43 Vassar Street, Cambridge, MA 02139-4307, USA., Groen IIA; Amsterdam Brain & Cognition (ABC) Center, University of Amsterdam, Nieuwe Achtergracht 129-B, 1018WS Amsterdam, the Netherlands; Video and Image Sense Lab, Informatics Institute, University of Amsterdam, LAB42, Science Park 900, 1098XH Amsterdam, the Netherlands., Cappaert N; Swammerdam Institute for Life Sciences, University of Amsterdam, Science Park 904, 1098XH Amsterdam, the Netherlands., Scholte HS; Department of Psychology, University of Amsterdam, Nieuwe Achtergracht 129-B, 1018WS Amsterdam, the Netherlands; Amsterdam Brain & Cognition (ABC) Center, University of Amsterdam, Nieuwe Achtergracht 129-B, 1018WS Amsterdam, the Netherlands. Electronic address: hsscholte@gmail.com.
Πηγή:	Current biology : CB [Curr Biol] 2026 Sep 21; Vol. 36 (18), pp. 4827-4834.e3. Date of Electronic Publication: 2026 Aug 31.
Τύπος έκδοσης:	Journal Article
Γλώσσα:	English
Στοιχεία περιοδικού:	Publisher: Cell Press Country of Publication: England NLM ID: 9107782 Publication Model: Print-Electronic Cited Medium: Internet ISSN: 1879-0445 (Electronic) Linking ISSN: 09609822 NLM ISO Abbreviation: Curr Biol Subsets: MEDLINE
Imprint Name(s):	Publication: Cambridge, MA : Cell Press Original Publication: London, UK : Current Biology Ltd., c1991-
Ιατρικοί όροι (MeSH):	Visual Cortex/physiology , Pattern Recognition, Visual/physiology , Visual Perception/physiology , Neural Networks, Computer, Humans ; Electroencephalography
Περίληψη:	Deep neural networks (DNNs) excel at predicting neural responses across the visual hierarchy,¹^,²^,³^,⁴^,⁵ a success widely interpreted as evidence of shared object recognition computations.⁶^,⁷ Yet improving DNN object recognition accuracy does not reliably increase neural predictivity,⁸^,⁹ and even untrained networks predict brain responses above chance.⁹^,¹⁰^,¹¹ This disconnect suggests that object recognition may not drive DNN-brain alignment. Texture-like statistics are represented in both DNNs and mid-level visual cortex (A.V. Jagadeesh and M. Livingstone, 2024, ICLR, presentation).¹²^,¹³^,¹⁴¹⁶ In natural images, these statistics are carried by objects and backgrounds, shaping representations and recognition in both systems.¹⁶^,¹⁷^,¹⁸^,¹⁹ Does DNN-brain alignment reflect a shared sensitivity to object-related information or texture-like statistics? To dissociate these factors, we recorded electroencephalograms (EEGs) from 57 participants viewing natural scenes, texture-synthesized images preserving local statistics while disrupting global form, and object-only images with backgrounds removed. If alignment reflects texture-like statistics, then it should peak for texture-synthesized images. If it reflects object-related processing, then alignment should be strongest for natural and object-only conditions, which preserve object information. We compared EEG responses with DNN activations via weighted representational similarity analysis.²⁰^,²¹ Texture-synthesized images yielded the strongest DNN-EEG alignment, peaking in early responses (<200 ms) and explaining up to ∼85% of noise-ceiling-normalized explainable variance versus ∼44% for natural and ∼55% for isolated objects. Crucially, object categories were more decodable for natural and object-only images than texture-synthesized images, yet these object-rich conditions showed weaker alignment. This dissociation reveals that DNNs capture the texture-statistical component of early visual responses while failing to explain later, object-related variance. (Copyright © 2026 Elsevier Inc. All rights reserved.)
Competing Interests:	Declaration of interests The authors declare no competing interests.
Contributed Indexing:	Keywords: DNN; EEG; RSA; deep neural networks; electroencephalography; feedforward processing; image statistics; noise ceiling; representational alignment; representational similarity analysis; texture bias; texture synthesis
Entry Date(s):	Date Created: 20260831 Date Completed: 20260921 Latest Revision: 20260921
Update Code:	20260922
DOI:	10.1016/j.cub.2026.08.008
PMID:	42673953
Βάση Δεδομένων:	MEDLINE

View record from ScienceDirect

Περιγραφή
ISSN:	1879-0445
DOI:	10.1016/j.cub.2026.08.008