Multi-Scale Dynamic Fixed-Point Quantization and Training for Deep Neural Networks | IEEE Conference Publication | IEEE Xplore