Quantizing Convolutional Neural Networks for Low-Power High-Throughput Inference Engines