Contexte/ Objectifs :

The research proposed in this project fall within the field of physical informed neural networks for electromagnetic applications. The aim is to use this type of method to generate synthetic data to deal with the lack of data in some applications. Until now, the solution has been to use simulators based on physical models. Several levels of modeling can be used such as approximate physical models or the so-called exact models. However, these models are either too simplified or too time-consuming and computationally demanding. Moreover, even if in theory the exact models can consider all the physical phenomena, at the end there always exist differences with real data.

The goal of this project is to evaluate the physical informed neural networks (PINNs) as an alternative solution to physical models. PINNs are becoming increasingly popular but there are still number research works to do, especially when dealing with 3D problems. We also must take into account the fact that the electromagnetic field is complex-valued. In this project we will first evaluate PINN to compute the scattered field from object considering 2D configurations. Depending on the results, 3D problems can be considered.

Profil du candidat :

We are looking for a highly motivated candidate to study PINNs for electromagnetic applications. The candidate should possess the following qualifications:

• A robust background in machine learning, signal processing, or applied mathematics

• Strong programming abilities in Python and PyTorch.

Formation et compétences requises :

Research engineer or Ph.D. with strong experience in deep learning and, if possible, PINNs. Basic knowledge of electromagnetism and modeling would be appreciated.