Hi everybody,

Recently I checked again the Word Models interactive paper. They use a to compress 64x64x3 images to 32 dim vectors. But on the site there is an interactive demo where I can set 15 values of the latent vector which will be decompressed. I guess these are the most out of the 32.

The question is, how can I rank the features by reconstruction variance (or I don̵;t really know how to call this)?

Source link
thanks you RSS link
( https://www.reddit.com/r//comments/9c1rla/d__the_most_important_features_at_the/)


Please enter your comment!
Please enter your name here