Plenary – “Topological Data Analysis and Deep Learning” (G. Carlsson)
Deep learning is a powerful methodology that permits the solution of many interesting problems in machine learning. However, it suffers from a few different problems.
It is often “data hungry”, it often doesn’t generalize well, it is subject to adversarial attacks, and it suffers from a general “opaqueness”.
In this talk we will discuss how topological methods can be used to mitigate these problems, with examples.