Cours de formation, événements et séminaires : Deep Learning Inversion of Borehole Resistivity Measurements. Part I: Choice of Loss
SÉMINAIRE GROUPE HYDROLOGIE DES EAUX SOUTERRAINES (UPC-CSIC)
Par : David Pardo (UPV/EHU, BCAM, et professeur de recherche d'Ikerbasque)
Jour : jeudi 17 octobre
Heure : à 12:15 h
Lieu : Département de génie civil et environnemental, salle de classe Modulo D2-CIHS, rez-de-chaussée
Abstract:
In geosteering operations, it is necessary to invert (interpret) borehole resistivity measurements in real-time. Recently, Deep Neural Networks (DNNs) have arisen as an alternative to perform such a complex task due to their unique ability to approximate different unknown functions.
In this presentation, we describe the application of interest and the need for real-time inversion. Then, we review basic concepts of DNNs when applied to inverse problems. Thereafter, we focus on the selection of proper loss functions for solving geosteering inverse problems. Some numerical results illustrate the performance and limitations of various loss functions.
Most derivations and ideas contained in the presentation are directly applicable to other inverse problems across different areas of engineering.