Deep Neural Network Compression with Single and Multiple Level Quantization