Informations générales
Intitulé de l'offre : Doctoral thesis on Energy efficient algorithms for Federated Learning (M/W) (H/F)
Référence : UPR8001-BALPRA-006
Nombre de Postes : 1
Lieu de travail : TOULOUSE
Date de publication : jeudi 7 septembre 2023
Type de contrat : CDD Doctorant/Contrat doctoral
Durée du contrat : 36 mois
Date de début de la thèse : 2 octobre 2023
Quotité de travail : Temps complet
Rémunération : 2 135,00 € gross monthly
Section(s) CN : Information sciences: bases of information technology, calculations, algorithms, representations, uses
Description du sujet de thèse
The growing interest in AI technologies has resulted in a demand for energy which is incompatible with the global objectives for sustainable development and climate change. While recent work has highlighted this incompatibility for classical learning algorithms, for emerging paradigms like Federated Learning (FL) there are no such studies. The aim of the ANR DELIGHT project is to provide a framework for evaluating the energy consumption of FL algorithms as well as to propose new algorithms with improved energy-efficiency. The proposed thesis topic fits in the second objective of
the DELIGHT project. Using techniques like Data Summarization, Gradient Compression, and Speed-scaling, the aim of the thesis is to exploit the heterogeneity of data in FL to reduce the energy footprint. Further questions related to participation of nodes, and the value they bring to the federation (including energy) will be also be investigated. The techiniques developed will be tested on tasks of computer vision and NLP using the toolkit Flower.
Contexte de travail
The work will be carried out at LAAS-CNRS
Le poste se situe dans un secteur relevant de la protection du potentiel scientifique et technique (PPST), et nécessite donc, conformément à la réglementation, que votre arrivée soit autorisée par l'autorité compétente du MESR.