Logan Riggs comments on How to Better Report Sparse Autoencoder Performance

[Logan Riggs]

I think some people use the loss when all features are set to zero, instead of strictly doing

I think this is an unfinished