'Artiphysiology' reveals V4-like shape tuning in a deep network trained for image classification
- 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.
-
BVLC reference caffenetBAIR BVLC CaffeNet Model.
Article and author information
Author details
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.
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.
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
-
- 2,937
- views
-
- 394
- downloads
-
- 64
- 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)
'Artiphysiology' reveals V4-like shape tuning in a deep network trained for image classification
eLife 7:e38242.
https://doi.org/10.7554/eLife.38242
Further reading
-
- Neuroscience
Techniques that enable precise manipulations of subsets of neurons in the fly central nervous system (CNS) have greatly facilitated our understanding of the neural basis of behavior. Split-GAL4 driver lines allow specific targeting of cell types in Drosophila melanogaster and other species. We describe here a collection of 3060 lines targeting a range of cell types in the adult Drosophila CNS and 1373 lines characterized in third-instar larvae. These tools enable functional, transcriptomic, and proteomic studies based on precise anatomical targeting. NeuronBridge and other search tools relate light microscopy images of these split-GAL4 lines to connectomes reconstructed from electron microscopy images. The collections are the result of screening over 77,000 split hemidriver combinations. Previously published and new lines are included, all validated for driver expression and curated for optimal cell-type specificity across diverse cell types. In addition to images and fly stocks for these well-characterized lines, we make available 300,000 new 3D images of other split-GAL4 lines.
-
- Neuroscience
Cell identification is an important yet difficult process in data analysis of biological images. Previously, we developed an automated cell identification method called CRF_ID and demonstrated its high performance in Caenorhabditis elegans whole-brain images (Chaudhary et al., 2021). However, because the method was optimized for whole-brain imaging, comparable performance could not be guaranteed for application in commonly used C. elegans multi-cell images that display a subpopulation of cells. Here, we present an advancement, CRF_ID 2.0, that expands the generalizability of the method to multi-cell imaging beyond whole-brain imaging. To illustrate the application of the advance, we show the characterization of CRF_ID 2.0 in multi-cell imaging and cell-specific gene expression analysis in C. elegans. This work demonstrates that high-accuracy automated cell annotation in multi-cell imaging can expedite cell identification and reduce its subjectivity in C. elegans and potentially other biological images of various origins.
