Chapter 14 - Visual neuroscience in the age of big data and artificial intelligence

Abstract

Significant progress in visual neuroscience has often followed somewhat serendipitous discoveries of specific stimulus preferences of individual neurons. For instance, the discovery of orientation-selectivity in the cat primary visual cortex, or face-selectivity in the macaque inferior temporal cortex, has shaped the nature of follow-up experiments and the field’s overall enthusiasm. These experiments have helped visual neuroscientists to qualitatively describe the coarse nature of representations in individual brain areas. However, arguably, their inferences often have limited scope with a narrow focus on the stimulus space that drives the activity of neurons within the respective brain areas of interest—causing a distraction from asking what mechanistic models might predict the neural responses in a brain area (or the animals’ behavior) given any possible input stimulus. In this chapter, I argue that despite the many insights gained from such traditional approaches, visual neuroscience is currently undergoing somewhat of a Renaissance. Spearheaded by the latest technological advances in graphics processing and our heightened ability to collect and efficiently store large-scale datasets, computer vision (CV) has produced very high performing image computable artificial neural network models. These models not only excel in CV tasks but also surprisingly serve as the current best models of the primate visual system, matching both its behavioral and neural outputs. However, despite mimicking the hierarchical architecture of the visual cortex and demonstrating more predictive power in explaining neural data compared to earlier models, these models are not yet perfect matches of the biological visual system. Future experimental questions in visual neuroscience can be driven and motivated by predictions, failures, and improvements of such models. Furthermore, this can serve as an inspiration and template for probing systems that support other brain functions.