Talk by Gunnar Carlsson (SMSS Colloquium)

Title: Topology for Artificial Intelligence

Abstract: Topological techniques (broadly construed) are turning out to be very useful in making sense of large data sets, and this fact has important consequences for artificial intelligence. The ideas involve the adaptation of homological methods as well as standard methods of homotopy theory to support the unsupervised analysis of complex data. I will discuss these methods, with examples of various kinds including the analysis of data sets arising out of deep learning, an extremely popular methodology for machine learning. We will also demonstrate that applications go beyond simple analysis of the algorithms but also permit the construction of architectures for efficient neural networks.