1 paper
J. Albericio, P. Judd, A. Delmás +2
We quantify a source of ineffectual computations when processing the multiplications of the convolutional layers in Deep Neural Networks (DNNs) and propose Pragmatic (PRA), an arch…