Quantization and Training of Neural Networks

Authors: Jacob et al. Year: 2017 ArXiv/Link: https://arxiv.org/abs/1810.08779

Summary

Foundations of quantization-aware training (QAT), enabling inference with low-precision fixed-point numbers.

Key Concepts

• Quantization-aware training
• Integer quantization
• Fixed-point
• Post-training calibration
• QAT

Impact

Established quantization as practical inference optimization

Category

Quantization

---