Problem: Using the MNIST handwritten data set, create and train an Autoencoder Neural Network that outputs both the image and a 90 degree rotation of the image. This will be a Y-shaped autoencoder, with a single encoder and two decoders.
Using both traditional DNN and CNN
For DNN, tried 0 degree alone, 90 degree, and the Y shape. Effect of the middle layer size are also tested. CNN only applied to 0 degree self in self out.