Learnable Companding Quantization for Accurate Low-bit Neural Networks | IEEE Conference Publication | IEEE Xplore