Introduction

1. Robust Quantization: One Model to Rule Them All
   paper

Use Kurtosis regularization (KURE) to force the weight distribution to uniform distribution

1. BSQ: EXPLORING BIT-LEVEL SPARSITY FOR MIXED-PRECISION NEURAL NETWORK QUANTIZATION
   paper

compose the quantization into bit-wise and optimize each bit

1. Ultra-Low Precision 4-bit Training of Deep Neural Networks
   paper