Opened this issue 7 years ago · 2 comments
AE with just one hidden layer seems to do a good job of dimensionality reduction.
Check with more layers
VAE with 2 latent values (encoded outputs dimension = 2) vs PCA with 2 principal eigenvalues.
