2019 Conference on Cognitive Computational Neuroscience

13-16 September 2019    Berlin, Germany

Technical Program

Paper Detail

Paper: PS-2B.20
Session: Poster Session 2B
Location: H Fläche 1.OG
Session Time: Sunday, September 15, 17:15 - 20:15
Presentation Time:Sunday, September 15, 17:15 - 20:15
Presentation: Poster
Publication: 2019 Conference on Cognitive Computational Neuroscience, 13-16 September 2019, Berlin, Germany
Paper Title: Visual Features for Invariant Coding by Face Selective Neurons
Manuscript:  Click here to view manuscript
License: Creative Commons License
This work is licensed under a Creative Commons Attribution 3.0 Unported License.
DOI: https://doi.org/10.32470/CCN.2019.1040-0
Authors: Wilbert Zarco, Winrich Freiwald, The Rockefeller University, United States
Abstract: Complex visual objects like faces are encoded through the primate ventral visual pathway in a network of cortical patches. Neurons across the nodes display specialized tuning to faces and increasing tolerance to image transformations. However, the exact features that neurons in different nodes use to attain selectivity and tolerance remain elusive. In this paper, we first quantified the representational content of neural populations in two fMRI-identified face patches to four attributes: viewpoint, identity, expression and mirror-symmetry. We found that neural population activity is driven by compartamentalized time-varying image attributes, and that multiple variables are represented in the anterior but not the posterior face patch. We then derived maps of feature selectivity by sampling images with Gaussian apertures that linked the evoked neuronal activity and the informative image features (IFs). This allowed us to evaluate the relationship between IFs and global stimulus tuning. We report that the set of discovered IFs explain the patterns of dissimilarity for the global viewpoint tuning. The alphabet of IFs also preserves local image preferences across changes in size and position. Crucially, the derived features are interpretable, and tend to cluster on consistent image regions, providing information about the global tuning that organize the neurons into functional groups.