논문리뷰: Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference

paper with code: https://paperswithcode.com/paper/quantization-and-training-of-neural-networks
CVPR 2018
Quantization, GeneralClassification
Dataset: ImageNet, COCO, ssd

📕 Summary

Background

co-design a training procedure to preserve end-to-end model accuracy post quantization

Problem (summarize)

CNN is not well suited for mobile deivces

Solution (This paper's keywork, key insight)

Algorithm

*depth-multipliers (DM)

Contribution