Model compression via distillation and quantization
2018·Arxiv