RePr: Improved Training of Convolutional Filters

Aaditya Prakash, James Storer, Dinei Florencio +1

A well-trained Convolutional Neural Network can easily be pruned without significant loss of performance. This is because of unnecessary overlap in the features captured by the net…