Principles of Learning in Distributed Brain Networks

Learning is a little-remarked miracle. With practice we can improve our performance on almost any task. However, there is a large gap between the remarkable learning abilities of organisms, and our theoretical understanding of how these abilities might arise. Even artificial machine learning systems, where we understand the basic components, are often black boxes into which we have little insight. The goal of this project is to use mathematical analyses to develop a fundamental understanding of learning dynamics in artificial neural networks, and to exploit this understanding to make predictions for experiments in psychology and neuroscience. In particular, this project studies how learning in any one brain area depends on how it is interconnected to all other brain areas, in order to understand how brain structure impacts learning. Greater insight into how we learn, beyond its inherent scientific significance, could lead to better machine learning systems, and improved educational interventions.