2019 Conference on Cognitive Computational Neuroscience

13-16 September 2019    Berlin, Germany

Technical Program

Paper Detail

Paper: PS-1A.64
Session: Poster Session 1A
Location: H Lichthof
Session Time: Saturday, September 14, 16:30 - 19:30
Presentation Time:Saturday, September 14, 16:30 - 19:30
Presentation: Poster
Publication: 2019 Conference on Cognitive Computational Neuroscience, 13-16 September 2019, Berlin, Germany
Paper Title: Does CNN Explain Tuning Properties of Macaque Face-Processing System?
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.2019.1063-0
Authors: Rajani Raman, Haruo Hosoya, Advanced Telecommunications Research Institute International, Japan
Abstract: Recent computational studies have emphasized quantitative similarity between convolutional neural networks (CNNs) and the visual ventral stream up to the primate inferotemporal (IT) cortex. However, whether such similarity holds for the face-selective areas, a subsystem of IT, is not clear. To address this question, we extensively investigate CNNs in terms of known tuning properties of the face-processing network in macaque IT. Specifically, we first trained an AlexNet-type CNN model with natural face images. Then, we conducted simulation of four physiological experiments (Freiwald, Tsao, & Livingstone, 2009; Freiwald & Tsao, 2010; Ohayon, Freiwald, & Tsao, 2012; Chang & Tsao, 2017) to make a correspondence between the model layers and the macaque face patches. As a result, we found that higher model layers explained well properties of anterior patches, while no layer had properties close to middle patches; this observation was consistent across model variation. Our results indicate that, although the near-goal representation of face-classifying CNNs has some similarity with the primate face processing system, the intermediate computational process might be rather different, thus calling for a more comprehensive model for better understanding of this system.