Study of the use of neural networks for high-performance calculations dedicated to the modeling of the transport of energy sources

Whatever their fields of application, modern computation codes are subject to requirements for speed of calculation and optimization of memory space occupation, which necessitate the use of advanced numerical methods. To meet these needs for massive data processing and intensive calculations, deep learning methods offer effective and extremely attractive alternative responses. This thesis manuscript presents a study on the coupling of artificial neural networks to high performance computing codes dedicated to simulations of complex physical phenomena, more particularly in connection with transport theories. The first application framework concerns the simulation of inertial confinement fusion experiments for the production of energy, and in particular, the modeling of nonlocal electronic heat transport. The second field of application is the transport and deposition of energy from particles in radiotherapy for the treatment of cancers. The two studies of the coupling of artificial neural networks to high performance computing codes developed by the IFCIA research team at the CELIA laboratory, are presented in this manuscript, and show encouraging results: for precision criteria largely sufficient for these applications, we obtain considerable time savings. This preliminary feasibility study encourages the CELIA team to continue this work in the years to come. (author)