'Artiphysiology' reveals V4-like shape tuning in a deep network trained for image classification

  1. Dean A Pospisil  Is a corresponding author
  2. Anitha Pasupathy
  3. Wyeth Bair
  1. University of Washington, United States

Abstract

Deep networks provide a potentially rich interconnection between neuroscientific and artificial approaches to understanding visual intelligence, but the relationship between artificial and neural representations of complex visual form has not been elucidated at the level of single-unit selectivity. Taking the approach of an electrophysiologist to characterizing single CNN units, we found many units exhibit translation-invariant boundary curvature selectivity approaching that of exemplar neurons in the primate mid-level visual area V4. For some V4-like units, particularly in middle layers, the natural images that drove them best were qualitatively consistent with selectivity for object boundaries. Our results identify a novel image-computable model for V4 boundary curvature selectivity and suggest that such a representation may begin to emerge within an artificial network trained for image categorization, even though boundary information was not provided during training. This raises the possibility that single-unit selectivity in CNNs will become a guide for understanding sensory cortex.

Data availability

No new datasets were generated in the course of this research. The model this research is based on is openly available from the Berkeley Artificial Intelligence Lab.

The following previously published data sets were used

Article and author information

Author details

  1. Dean A Pospisil

    Department of Biological Structure, University of Washington, Seattle, United States
    For correspondence
    deanp3@uw.edu
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-5793-2517
  2. Anitha Pasupathy

    Department of Biological Structure, University of Washington, Seattle, United States
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3808-8063
  3. Wyeth Bair

    Department of Biological Structure, University of Washington, Seattle, United States
    Competing interests
    The authors declare that no competing interests exist.

Funding

National Science Foundation (Graduate Research Fellowship)

  • Dean A Pospisil

National Science Foundation (CRCNS Grant IIS-1309725)

  • Anitha Pasupathy
  • Wyeth Bair

Google (Google Faculty Research Award)

  • Wyeth Bair

National Institutes of Health (Grant R01 EY-018839)

  • Anitha Pasupathy

National Institutes of Health Office of Research Infrastructure Programs (Grant RR-00166 to the Washington National Primate Research Center)

  • Anitha Pasupathy

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Reviewing Editor

  1. Eilon Vaadia, The Hebrew University of Jerusalem, Israel

Ethics

Animal experimentation: All animal procedures for this study, including implants, surgeries and behavioral training, conformed to NIH and USDA guidelines and were performed under an institutionally approved protocol at the Johns Hopkins University (Pasupathy and Connor, 2001) protocol #PR98A63 and the University of Washington (El-Shamayleh and Pasupathy, 2016) UW protocol #4133-01.

Version history

  1. Received: May 14, 2018
  2. Accepted: December 17, 2018
  3. Accepted Manuscript published: December 20, 2018 (version 1)
  4. Version of Record published: January 16, 2019 (version 2)
  5. Version of Record updated: February 1, 2021 (version 3)

Copyright

© 2018, Pospisil et al.

This article is distributed under the terms of the Creative Commons Attribution License permitting unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 2,787
    views
  • 379
    downloads
  • 58
    citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Dean A Pospisil
  2. Anitha Pasupathy
  3. Wyeth Bair
(2018)
'Artiphysiology' reveals V4-like shape tuning in a deep network trained for image classification
eLife 7:e38242.
https://doi.org/10.7554/eLife.38242

Share this article

https://doi.org/10.7554/eLife.38242

Further reading

    1. Neuroscience
    Ya-Hui Lin, Li-Wen Wang ... Li-An Chu
    Research Article

    Tissue-clearing and labeling techniques have revolutionized brain-wide imaging and analysis, yet their application to clinical formalin-fixed paraffin-embedded (FFPE) blocks remains challenging. We introduce HIF-Clear, a novel method for efficiently clearing and labeling centimeter-thick FFPE specimens using elevated temperature and concentrated detergents. HIF-Clear with multi-round immunolabeling reveals neuron circuitry regulating multiple neurotransmitter systems in a whole FFPE mouse brain and is able to be used as the evaluation of disease treatment efficiency. HIF-Clear also supports expansion microscopy and can be performed on a non-sectioned 15-year-old FFPE specimen, as well as a 3-month formalin-fixed mouse brain. Thus, HIF-Clear represents a feasible approach for researching archived FFPE specimens for future neuroscientific and 3D neuropathological analyses.

    1. Neuroscience
    Amanda Chu, Nicholas T Gordon ... Michael A McDannald
    Research Article

    Pavlovian fear conditioning has been extensively used to study the behavioral and neural basis of defensive systems. In a typical procedure, a cue is paired with foot shock, and subsequent cue presentation elicits freezing, a behavior theoretically linked to predator detection. Studies have since shown a fear conditioned cue can elicit locomotion, a behavior that - in addition to jumping, and rearing - is theoretically linked to imminent or occurring predation. A criticism of studies observing fear conditioned cue-elicited locomotion is that responding is non-associative. We gave rats Pavlovian fear discrimination over a baseline of reward seeking. TTL-triggered cameras captured 5 behavior frames/s around cue presentation. Experiment 1 examined the emergence of danger-specific behaviors over fear acquisition. Experiment 2 examined the expression of danger-specific behaviors in fear extinction. In total, we scored 112,000 frames for nine discrete behavior categories. Temporal ethograms show that during acquisition, a fear conditioned cue suppresses reward seeking and elicits freezing, but also elicits locomotion, jumping, and rearing - all of which are maximal when foot shock is imminent. During extinction, a fear conditioned cue most prominently suppresses reward seeking, and elicits locomotion that is timed to shock delivery. The independent expression of these behaviors in both experiments reveal a fear conditioned cue to orchestrate a temporally organized suite of behaviors.