MNIST_latent_space_exploring

we project the MNIST dataset to 2D latent space using convolutional autoencoder

convolutional autoencoder Model architecture

encoder architecture as follow

decoder architecture as follow

Training

we trained the ae model using adam optimizer and the SSIM as loss function

Results

traversing latent space video link

ploting of the projected Training data

ploting of the projected testset

entropy of the latent space

END