Stagiaire R&D Intelligence Artificielle

Job description

PALAISEAU-IPVF(FRA)
Description du poste

Contexte et Environnement du poste

Multiple Causal Inference with Application to Welding Data AI – Research Topics Today’s machine learning applications are largely based on identifying association patterns to achieve high predictive accuracy.
However, most of the time, accuracy of such prediction is not the primary goal in the industry setting. Answering causal questions from data is often key to discover how to alter and optimize the system of interest and to prevent system failure.
For this reason, the machine learning community’s interest in causality has significantly increased in recent years [1, 2]. Traditionally, causal inference relies on the potential outcome to study how a single cause affects an outcome [3]. However, many engineering applications involve multiple causes and thus require to simultaneously study the effect of different causes that can affect an outcome of interest.

In this context, Total is investigating the use of the recently developed deconfounder algorithm [4, 5] that infers a latent variable as a substitute for unobserved confounders and then uses that substitute to perform causal inference. This very interesting approach has, to our knowledge, not yet been applied to industrial settings. Identified Industrial Application The research focus of the internship will be paired with an industrial application.
The multi-causes approach will be tested on welding data to optimize their yield strength. It is important to notice that welding has a key role in many industrial developments. Most of the Group's investments are indeed dependent on the success of welding operations:
·  LNG plan - Offshore wind power (fixed or floating) - Refineries and existing pipeline network
·  New pipeline network for bio-gas
·  Retrofitting for addition of hydrogen to the existing network … The impact of "non-quality" in welding is also very significant, either because of the risk of catastrophic failure, or in terms of project delay if the non-conformities are discovered before the start-up of the installations. Parameters influencing the quality and productivity of welding operations are multiple, overlapping (sometimes dependent on each other), uncertain (due to measurement tolerances and variability intrinsic to metallurgy). Therefore, existing physic-based software’s are particularly expensive and unable to model the complex physical phenomena controlling the welding strength. In this basis, causal inference is a promising alternative method to overcome the existing modeling limitations. References 1 Pearl, J. The Seven Tools of Causal Inference, with Reflections on Machine Learning Communications of the ACM, 2019 2 Schölkopf, V. Causality for Machine Learning. https://arxiv.org/pdf/1911.10500.pdf 3 Rubin D Causal Inference Using Potential Outcomes: Design, Modeling, Decisions Journal of the American Statistical Association 2005 4 Wang and Blei. The Blessings of Multiple Causes https://arxiv.org/pdf/1805.06826.pdf 5 D’Amour, A. (2019). On multi-cause causal inference with unobserved confounding: Counterexamples, impossibility, and alternatives. arXiv preprint arXiv:1902.10286

Objectifs, Missions et Activités

The responsibilities of the intern are twofold:
·  Evaluation of existing multiple causes methods (hypothesis on the data, limitations …)
·  Application to industrial data (development of a code/notebook, documentation, presentation of the method to data science and engineering team) The intern will be part of the transverse computational and data science R&D program. The internship is based in Palaiseau, starting in Winter/Spring 2021 (4 to 6 months). Possibility of scientific publications or presentations to technical conferences. Knowledge in R/Python required.

During this internship you will the occasion to develop your skills/knowledge in A.I. , Causal Inference & Statistics
Réfèrence

37430BR
Pays

France
Durée du contrat

6
A propos de nous/Profil de l'entreprise

BETTER ENERGY NEEDS YOU

Donnez le meilleur de vous-même à l’énergie ! Rejoignez TOTAL : plus de 500 métiers différents dans 130 pays. Une entreprise responsable avec des standards de sécurité et d’éthique forts, des perspectives d’évolution de carrière variées, une culture de l’innovation et une mission partagée par les 100.000 collaborateurs du Groupe : rendre l’énergie meilleure jour après jour.

Desired profile

Profil recherché

Connaissances techniques et informatiques indispensables

R/Python, Statistical/ Machine Learning
You have skills in coding and in scientific writing and presenting

Niveau d’étude actuel

You are currently following an engineer course (3rd year) or a M2 and you are looking for a 6month internship 
(Start : 01/03)
Métier

Resp Générales Rch et Dév, R&D Applications / Systèmes
Intitulé du poste

Stagiaire R&D Intelligence Artificielle
Unité de temps

Mois
Branche

Exploration and Production
Région, département, localité

91 - Essonne
Domaine professionnel

Recherche Innovation&Développt
Niveau d’expérience requis

Moins de 3 ans
Type d’emploi

Stage conventionné