Code for a cancelled project/paper. Written in theano during Spring 2017. Here, instead of minimizing f(A) where AA^T=I, we minimize f((BB^T)^-1/2 B). Let A_hat = (BB^T)^-1/2 B, we can see that A_hat is always orthogonal if B is full-row rank. Gradient w.r.t B is
-
Notifications
You must be signed in to change notification settings - Fork 0
e13000/another_orthogonal_rnn
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
About
Orthogonal rnn using inline orthogonalization, written in theano.
Resources
Stars
Watchers
Forks
Releases
No releases published
Packages 0
No packages published