I'll Know It When I See It: Bridging Perception and Conceptual Learning in Minds and Machines

Abstract

Recognising and understanding what we see is essential throughout life, from infancy to adulthood. While advances in brain imaging and computational models have shed light on how the adult brain processes visual information, much remains unknown about how this ability develops in infants. This thesis explores how conceptual knowledge emerges from perceptual input in the brain during infancy, and how these insights can improve artificial intelligence. Firstly, I examined what can be learned about objects from our everyday surroundings using an infant-inspired computational model. Then, I studied how the adult brain responds to patterns in the things we see with functional MRI. Finally, the developmental origins of visual recognition were explored in a novel awake neuroimaging study of infants at 2 and 9 months old (n=130), finding evidence for both innate and learned components. I will discuss the potential of this methodological innovation for developmental cognitive computational neuroscience, and the empirical and theoretical advances for biological and computational vision.