Once applied dropout is applied ,how to keep the same structure for applying on the gradient (tensorflow) -
here do:
i have nn (let's g(x))
i can output g(x) drop out (it drop out number of neurons) s fine me.
but right after, i'd compute gradient of same g(x) (same hidden units dropped out).
is there way memorize units dropped out , apply them gradient in tensorflow?
Comments
Post a Comment