Studying infant visual cortex in the scanner and in silico

Abstract:

How does the human brain begin to make sense of the visual world? Developmental science has long explored this question through infants’ looking behavior, offering insights into early perception. Yet, the neural processes that give rise to these behaviors remain unclear. Recent advances in awake infant fMRI now allow us to measure neural activity in babies as they engage with rich visual stimuli, opening a new window onto the origins of cognition. Using data from 130 infants scanned at two months of age, we find that category structure is already present in high-level visual cortex at this early stage. To interpret the features underlying these representations, we apply deep neural network (DNN) models of vision and find that networks trained on infant-like visual experience best capture the geometry of infants’ brain responses. Together, these findings demonstrate how combining neuroimaging with computational modeling can illuminate the earliest building blocks of human visual understanding. At the same time, they highlight the unique challenges of studying the developing brain — from the complexity of data collection to the interpretation of rapidly changing neural systems.
 

Speaker Bio

Clíona is a postdoctoral researcher at Stanford University, working with Cameron Ellis and Dan Yamins. She earned her PhD from Trinity College Dublin, where she worked on awake infant fMRI and methods for modeling infant vision. At Stanford, Clíona studies how human infants learn so efficiently, how this learning can be modeled with the latest advances in AI, and how these combined approaches inform our understanding of the developing mind.