Predicting Parameters in Deep Learning

Misha Denil, Babak Shakibi, Laurent Dinh +2

We demonstrate that there is significant redundancy in the parameterization of several deep learning models. Given only a few weight values for each feature it is possible to accur…