Lorenzo Crocco, Research Director at CNR - IREA, will give a seminar "Physics-assisted Deep Learning for Electromagnetic Inverse Scattering" on Monday October 30, at 11:00 a.m., at Institut Fresnel, Pierre Cotton room.
Abstract : Deep learning is nowadays becoming a ubiquitous paradigm in applied science, thanks to its capability of providing extremely powerful tools to successfully solve complex classification or regression tasks. On the other hand, given its data-driven nature, deep learning performs well in contexts where data are abundant. When this is not the case, alternative strategies are needed to overcome the poor outcome of the learning stage due to the lack of data. In physics-assisted approaches, this issue is coped with incorporating specific knowledge on the problem at hand into either the inputs or the internal structure of the deep learning. architecture, so that the learning task is assisted and to some extent simplified by the physics of the problem. In electromagnetic inverse scattering problems, such an approach appears to be very convenient, as it allows to exploit the broad knowledge on the problem (e.g., the spatial properties of scattered fields) as well as reliably preprocess raw data using effective inversion approaches (e.g. qualitative inversion methods). In the talk, I will discuss some possible ways to exploit physics-assisted deep learning for solving electromagnetic inverse scattering problems and give some examples relate to microwave imaging applications.
CNR : Consiglio Nazionale delle Ricerche
IREA : Institute for Electromagnetic Sensing of the Environment
See here for more info on his research activities :https://www.researchgate.net/profile/Lorenzo-Crocco
ResearchGate
Flux RSS