Deep Learning on MNIST with causal model
Content
In this note, I make a back-up of the process and result of train MNIST with pyro and pytorch.
The MNIST dataset comprised of images of digits and their digit "labels", and there is a related dataset called Typeface MNIST in which the digits are illustrated with different fonts. Each image of digit is featured with digit "labels" (0~9) and its type "labels" (handwritten = 1, not handwritten = 0). According to the author of Causal AI, we could build a causal generative model with the DAG.