Topological methods for Artificial Intelligence (PIMS network-wide colloquium)

The recent striking developments in artificial intelligence underscore the idea that we need to (a) understand better what goes on inside neural networks, (b) thereby gain more control over them and (c) develop methods that are less expensive and apply directly to specialized situations.  In this talk, I will talk about work on methods using topological methods to address these three problems.  In particular, I will include numerous examples.