UMR 5672

logo de l'ENS de Lyon
logo du CNRS
You are here: Home / Seminars / Machine Learning and Signal Processing / Analysis of complex physical systems with information theory and statistical learning

Analysis of complex physical systems with information theory and statistical learning

Léonard Seydoux (post-doctorant ISTerre, Grenoble)
When Mar 02, 2021
from 10:00 to 11:00
Attendees Léonard Seydoux
Add event to calendar vCal
iCal

Title : Analysis of complex physical systems with information theory and statistical learning

Asbtract : Characterizing complex systems where many intricate physical processes are acting is a challenge that I propose to address with the analysis of large datasets. While physical models can describe the behavior of such systems in a broad sense, complex systems may depend on many disregarded variables, and therefore may not be accurately described by these general models. Statistical learning helps to build models that describe with a higher accuracy these complex systems, taking into account the contextual properties of each system individually, with the drawback of being poorly interpretable and having a low potential for generalization. My research focuses on building hybrid models based on statistical learning and information theory, constrained by the a priori knowledge on the underlying physics. Following the main idea of statistical learning, I work on learning relevant representations of multi-scale and multi-physics data to solve a problem of interest (detection, imaging, monitoring). With a focus on wave physics in complex media, I will show that we can improve the characterization of random wavefields for source detection, separation and location, and structural imaging that can find application in various domains. I will illustrate the different methods with examples from geophysics data (seismic, geodesic) in volcanic and tectonic settings, lab experiments on metamaterials, and medical imaging.  

En savoir plus :  https://scholar.google.fr/citations?hl=fr&user=TqLdn9YAAAAJ&view_op=list_works&sortby=pubdate

Séminaire en ligne : https://lpensl.my.webex.com/ 
Room : Webconf3