Tactic: Use Quantization-Aware Training

Tactic sort: Awesome Tactic

Type: Architectural Tactic

Tags: machine-learning model-training

Title

Use Quantization-Aware Training

Description

Quantization-aware training is a technique used to train neural networks to convert data types to lower precision. The idea is to use fixed-point or integer representations instead of the more commonly used higher-precision floating-point representations. This improves the performance and energy efficiency of the model in federated learning.

Participant

Data Scientist

Related software artifact

Model

Context

Machine Learning

Software feature

Model Training

Tactic intent

Improve energy efficiency by using quantization-aware training to convert high-precision data types to lower precision

Target quality attribute

Accuracy

Other related quality attributes

Energy Efficiency

Measured impact

< unknown >

Source

Minsu Kim, Walid Saad, Mohammad Mozaffari, and Merouane Debbah. 2021. On the Tradeoff between Energy, Precision, and Accuracy in Federated Quantized Neural Networks. In ICC 2022 - IEEE International Conference on Communications. 2194–2199. [DOI](https://doi.org/10.1109/ICC45855.2022.9838362); Martino Sorbaro, Qian Liu, Massimo Bortone, and Sadique Sheik. 2020. Optimizing the Energy Consumption of Spiking Neural Networks for Neuromorphic Applications. Frontiers in Neuroscience 14 (2020), 662. [DOI](https://doi.org/10.3389/fnins.2020.00662)