Dropout BW does not work
Benny-Nottonson opened this issue · comments
Describe the bug
The backwards gradient for dropout does not work since the behavior is technically random, find a way around this
Expected behavior
The bw gradient of the Dropout function should be trainable, need to do research on how