We will talk about connections between topology and machine learning. Statistics and local geometry has for a long time served as the underpinning of machine learning, but recent developments allow us to incorporate global geometry and topology with great results. First, topological features such as persistent homology and clustering methods can help us understand what neural networks learn and construct topology inspired learning theory. Next, many of these features are differentiable which allows us to create a general purpose machine learning layer that allows anyone to easily incorporate topology into any machine learning pipeline. Lastly, we will talk about universal function approximation on graphs and the promise of unsupervised learning and pre-training.