bench sweep
I've long suspected that there was smth wrong w backprop
Created on August 12|Last edited on August 15
Comment
does any backprop work?
Computing group metrics from first 36 groups
plast_clip, updater
plast_clip set to 1e4, updater vs. recurrence. We hope to see solid performance on both dfa and backprop even without recurrence.
is clip_weights useful at all, and if so, what value of it?
is residual_connections a useful parameter?
Add a comment