Closed JohnGiorgi closed 5 years ago
A new optimizer, Yogi was introduced at NIPS. Authors claim better performance than existing adaptive gradient methods (e.g. RMSProp, Adam, Adadelta).
This has already been implemented in keras-contrib here.
This would be an extremely quick thing to try. If it improves performance we could swap it in for Nadam.
A new optimizer, Yogi was introduced at NIPS. Authors claim better performance than existing adaptive gradient methods (e.g. RMSProp, Adam, Adadelta).
This has already been implemented in keras-contrib here.
This would be an extremely quick thing to try. If it improves performance we could swap it in for Nadam.