REX - Continuous estimation of the environmental impact of Machine Learning solutions : methodology, best practices and tools

Présentation de l'intervenant

• Samuel Rincé, Lead Data Scientist at Alygne & OS Contributor at Boavizta

Introduction

The environmental impact of AI solutions can be significant and is set to grow in the coming years. Data scientists need to be aware of this, measure the environmental impact of their solutions, and try to reduce it.

Even if it is impossible to precisely measure the environmental impact (energy consumption, raw materials, greenhouse gas emissions, water consumption, etc.) throughout the life cycle of an AI solution (Life Cycle Assessment), there are tools available to estimate this impact, notably the energy consumption, or even hardware production and use, regarding greenhouse gas emissions.

Among many other tools, we can mention:

• for the development phase: Code Carbon, Zeus
• for the validation phase: SoftAware
• for the prouction phase: Scaphandre

Code Carbon monitors energy consumption, converted into greenhouse gas emissions, when your code is executed.

Zeus, monitors energy consumption and optimizes energy consumption during model training.

SoftAware, monitors energy consumption throughout the CI pipeline.

Scaphandre, monitors the energy consumption of your applications in production on a server.

To find out more about this topic and the tools, you can view the slides and replay on our youtube channel. And don't forget to check on our future workshops !