How can we use the latent manifolds of image data in the structure of convolutional neural networks? Topological data analysis is a tool used to analyze the internal states of a neural network, and this paper leverages that to better analyze image and video data.

• How can we use the latent manifolds of image data in the structure of convolutional neural networks?
• How are image patch data and the Klein bottle connected?
• How can we use topology to understand the internal states of a CNN?