CATERPILLAR: Coarse Grain Reconfigurable Architecture for accelerating the training of Deep Neural Networks | IEEE Conference Publication | IEEE Xplore