Date : Mardi 18 février 2025, 14 heures.
Lieu : Lagrange Gris (L101), Inria Sophia Antipolis.
Titre : Green Federated Learning: Scheduling clients’ participation to reduce the training carbon footprint.
Résumé : With the growing adoption of deep learning models in everyday applications, reducing their carbon footprint has become increasingly crucial, as training and deployment consume significant computational resources, leading to higher energy demands.
This work investigates Green Federated Learning (GreenFL), proposing an optimized client participation scheduling strategy based on the carbon emission values of power sources.
Given that client data is distributed in a non-IID (non-independent and identically distributed) fashion, we explore whether such a scheduling approach can still yield models with high performance while minimizing environmental impact.
