Poster
Predicting Parameters in Deep Learning
Misha Denil · Babak Shakibi · Laurent Dinh · Marc'Aurelio Ranzato · Nando de Freitas

Fri Dec 6th 07:00 -- 11:59 PM @ Harrah's Special Events Center, 2nd Floor #None

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 accurately predict the remaining values. Moreover, we show that not only can the parameter values be predicted, but many of them need not be learned at all. We train several different architectures by learning only a small number of weights and predicting the rest. In the best case we are able to predict more than 95% of the weights of a network without any drop in accuracy.

Author Information

Misha Denil (University of Oxford)
Babak Shakibi (UBC)
Laurent Dinh (École Centrale Paris)
Marc'Aurelio Ranzato (Facebook AI Research)
Nando de Freitas (UBC)

More from the Same Authors