Geoffrey Goodhill, Professor, Queensland Brain Institute
Learning about learning in the brain using mathematical and statistical models
Abstract: As the brain develops it adapts to the statistical structure of the environment. While supervised learning algorithms such as deep learning have proven very powerful for practical applications, we still have much to discover about the unsupervised methods that drive early adaptation and learning in real brains. First, I will discuss work that suggests how maps of visual features in the mammalian cortex may develop according to an optimisation principle that can be mapped onto the traveling salesman problem. This successfully predicts how maps of ocular dominance and orientation selectivity change in response to altered environmental input. Second I will discuss our more recent work looking at the early development of neural activity in the zebrafish brain, and how this is related to the development of the fish's behavior. This includes developing a community detection algorithm for identifying groups of neurons that fire together, a Hebbian model for understanding how such groups might form, and a latent variable model for separating evoked from spontaneous activity. In the long term we hope that understanding more about the computational principles driving neural development will help inspire new advances in artificial intelligence.
If you are a member of the WashU community, login with your WUSTL Key to interact with events, personalize your calendar, and get recommendations.Login with WUSTL Key
If you are not a member of the WashU community, please login via one of the options below to interact with our calendar.