Văn bản này: Speeding up convolutional neural networks with low rank expansions