'Artiphysiology' reveals V4-like shape tuning in a deep network trained for image classification
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....
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
eLife Sciences Publications Ltd
2018-12-01
|
| Series: | eLife |
| Subjects: | |
| Online Access: | https://elifesciences.org/articles/38242 |
| _version_ | 1811227848493498368 |
|---|---|
| author | Dean A Pospisil Anitha Pasupathy Wyeth Bair |
| author_facet | Dean A Pospisil Anitha Pasupathy Wyeth Bair |
| author_sort | Dean A Pospisil |
| collection | DOAJ |
| description | 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. |
| first_indexed | 2024-04-12T09:48:55Z |
| format | Article |
| id | doaj.art-8981d629742d446a977f1e0fb3c86cbd |
| institution | Directory Open Access Journal |
| issn | 2050-084X |
| language | English |
| last_indexed | 2024-04-12T09:48:55Z |
| publishDate | 2018-12-01 |
| publisher | eLife Sciences Publications Ltd |
| record_format | Article |
| series | eLife |
| spelling | doaj.art-8981d629742d446a977f1e0fb3c86cbd2022-12-22T03:37:53ZengeLife Sciences Publications LtdeLife2050-084X2018-12-01710.7554/eLife.38242'Artiphysiology' reveals V4-like shape tuning in a deep network trained for image classificationDean A Pospisil0https://orcid.org/0000-0002-5793-2517Anitha Pasupathy1https://orcid.org/0000-0003-3808-8063Wyeth Bair2Department of Biological Structure, Washington National Primate Research Center, University of Washington, Seattle, United StatesDepartment of Biological Structure, Washington National Primate Research Center, University of Washington, Seattle, United States; University of Washington Institute for Neuroengineering, Seattle, United StatesDepartment of Biological Structure, Washington National Primate Research Center, University of Washington, Seattle, United States; University of Washington Institute for Neuroengineering, Seattle, United States; Computational Neuroscience Center, University of Washington, Seattle, United StatesDeep 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.https://elifesciences.org/articles/38242neural networkV4shape perceptionventral stream |
| spellingShingle | Dean A Pospisil Anitha Pasupathy Wyeth Bair 'Artiphysiology' reveals V4-like shape tuning in a deep network trained for image classification eLife neural network V4 shape perception ventral stream |
| title | 'Artiphysiology' reveals V4-like shape tuning in a deep network trained for image classification |
| title_full | 'Artiphysiology' reveals V4-like shape tuning in a deep network trained for image classification |
| title_fullStr | 'Artiphysiology' reveals V4-like shape tuning in a deep network trained for image classification |
| title_full_unstemmed | 'Artiphysiology' reveals V4-like shape tuning in a deep network trained for image classification |
| title_short | 'Artiphysiology' reveals V4-like shape tuning in a deep network trained for image classification |
| title_sort | artiphysiology reveals v4 like shape tuning in a deep network trained for image classification |
| topic | neural network V4 shape perception ventral stream |
| url | https://elifesciences.org/articles/38242 |
| work_keys_str_mv | AT deanapospisil artiphysiologyrevealsv4likeshapetuninginadeepnetworktrainedforimageclassification AT anithapasupathy artiphysiologyrevealsv4likeshapetuninginadeepnetworktrainedforimageclassification AT wyethbair artiphysiologyrevealsv4likeshapetuninginadeepnetworktrainedforimageclassification |