Compression of Deep Convolutional Neural Networks for Efficient Inference on Vector Processors