AUTHOR=Celeghin Alessia , Borriero Alessio , Orsenigo Davide , Diano Matteo , Méndez Guerrero Carlos Andrés , Perotti Alan , Petri Giovanni , Tamietto Marco 

TITLE=Convolutional neural networks for vision neuroscience: significance, developments, and outstanding issues

JOURNAL=Frontiers in Computational Neuroscience

VOLUME=Volume 17 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2023.1153572

DOI=10.3389/fncom.2023.1153572

ISSN=1662-5188

ABSTRACT=Convolutional Neural Networks (CNN) are computational models inspired by biological brains' architecture and basic functions. They can attain human-like performance in several tasks by learning from experience, thus enabling empirical exploration of how visual functions and neural representations may originate in the real brain from a limited set of computational principles. After considering the basic principles of DNNs, we discuss the opportunities and challenges of endorsing DNNs as in silico models of the primate visual system. Specifically, we outline several emerging notions about the anatomical and physiological properties of the visual system that still need to be incorporated systematically in current CNN models. These tenets include the implementation of parallel processing routes from the early stages of retinal input and the reconsideration of several assumptions concerning the serial progression of information flow. We propose examples of design choices and architectural constraints that may permit a closer match to biology and provide causal evidence of the predictive link between the artificial and biological visual brain. This principled perspective would potentially lead to new research questions and applications of CNNs beyond modelling object recognition in the ventral visual stream.