Fetching the paper…

Additive Powers-of-Two Quantization: An Efficient Non-uniform Discretization for Neural Networks · Around