# Quantifying the Carbon Emissions of Machine Learning

**Authors:** Alexandre Lacoste, Alexandra Luccioni, Victor Schmidt, Thomas Dandres  
**arXiv:** [1910.09700](https://arxiv.org/abs/1910.09700)  
**Date:** 2019-10-21  
**Summary page:** https://www.summarizepaper.com/en/paper/1910.09700v2/

## Key points

- Machine Learning has a growing impact on the environment
- Several factors contribute to carbon emissions during neural network training, including server location, energy grid, training duration, and hardware make/model
- Researchers have developed an ML Emissions Calculator tool to approximate carbon emissions based on these variables
- The tool takes into account publicly available data for energy consumption, compute regions, CO2eq emissions per kWh and potential offsets bought by providers
- Individuals and organizations can take actionable steps to mitigate their carbon impact such as choosing more efficient hardware or quantifying their emissions
- The research team intends to maintain an open and transparent approach by updating data as more information becomes available
- This paper serves as a call-to-action for ML practitioners to be mindful of their environmental impact and take steps towards reducing it.

## Summary

The impact of Machine Learning on the environment is a growing concern and there are several crucial factors that contribute to the carbon emissions produced during the training of neural networks. These include the location of the server used for training, the energy grid it uses, the length of the training procedure and even the make and model of hardware used. To address this issue, researchers have developed an ML Emissions Calculator tool that allows practitioners to approximate their carbon emissions based on these variables. The tool takes into account publicly available data for energy consumption, compute regions, CO2eq emissions per kWh and potential offsets bought by providers. Additionally, there are actionable items that individuals and organizations can adopt to mitigate their carbon impact such as choosing more efficient hardware or quantifying their emissions. The research team behind this work intends to maintain an open and transparent approach by updating data as more information becomes available. This paper serves as a call-to-action for ML practitioners to be mindful of their environmental impact and take steps towards reducing it.