Merge branch 'job_3d31d5b96bdff55321c73093f851dc8a' into 'master'

New job offer job_3d31d5b96bdff55321c73093f851dc8a

See merge request !467

content/job_offers/job_3d31d5b96bdff55321c73093f851dc8a.md

+Title: Artificial Intelligence algorithms applied to gyro-kinetic codes
+Date: 2019-12-20 14:38
+Slug: job_3d31d5b96bdff55321c73093f851dc8a
+Category: job
+Authors: David Zarzoso
+Email: david.ZARZOSO-FERNANDEZ@univ-amu.fr
+Job_Type: Stage
+Tags: stage
+Template: job_offer
+Job_Location: Marseille
+Job_Duration: 5 mois
+Job_Website: 
+Job_Employer: Aix-Marseille university
+Expiration_Date: 2020-03-13
+Attachment: 
+
+The proposed internship is centred around nuclear fusion in magnetically confined plasmas. Such plasmas are heated up to hundreds million degrees and exhibit strong gradients, representing therefore intrinsically complex systems where many instabilities can develop on disparate spatio-temporal scales. Such instabilities are deleterious for the good confinement of energy and particles and they must absolutely be understood, predicted and eventually controlled. For this purpose, modelling the plasma is one of the techniques currently employed. In this internship we will focus on the gyro-kinetic approach, which consists in reducing the dimensionality of the particle phase-space from 6D (3 in velocity and 3 in real space) down to 5D (2 in velocity and 3 in real space) by averaging over the fastest gyro-motion. Even though such approach represents a simplification, it remains extremely expensive in terms of computational cost when modelling fusion plasmas accounting for all the scales characterizing the dynamics. This advocates for new methods in order to minimise the cost and the time to predict the behaviour of future plasma scenarios such as in ITER. In this context, Artificial Intelligence (AI) algorithms are one of the candidates and constitute the central part of the internship. The candidate will use a simplified gyro-kinetic code, where the trajectories of the particles will be integrated in the presence of an imposed electro-magnetic field. Different algorithms based on unsupervised and supervised machine learning will be applied. The goal is twofold. First, use unsupervised algorithms such as clustering techniques to classify particles depending on the physics at play. Second, apply supervised algorithms using deep neural networks to detect rare events such as avalanches of particles towards the edge and other nonlinear phenomena.
+The internship will be jointly supervised by David Zarzoso (CNRS, PIIM, Marseille) and Virginie Grandgirard (CEA, Cadarache).