Journée(s) Machine Learning et Physique Nucléaire

29 oct. 2019, 09:00 → 30 oct. 2019, 17:00 Europe/Paris

Bâtiment 108, 1er étage (Orsay)

L'objectif de cette réunion est d'informer, d'échanger, de débattre sur l'application des technologies relevant du "Machine Learning" en privilégiant des utilisations en lien avec la Physique Nucléaire.

La réunion sera découpée en quatre sessions :

1. Informations générales 
   • Panorama, possibilités, enjeux et technologies seront présentés
2. Example(s) d'application(s) concrète(s)
   • Objectif: montrer une(des) technologie(s) mises en oeuvre
3. Retours / attentes de la communauté 
   • Objectif: collecter les besoins (actions) de la communauté, donner des pistes pour commencer ...
4. Table ronde : discussions 

 

*** Appel à contribution fermé ***

Si vous souhaitez contribuer, notamment lors des sessions 2 ou 3, n'hésitez pas à nous envoyer vos contributions (meme si le travail sur les aspects Machine Learning n'a pas encore commencé) en donnant quelques lignes d'explications et éventuellement avec un titre provisoire

Adrien, Eric et Olivier

 

The goal of this meeting is to inform, exchange, discuss on Machine Learning technologies in particular for applications in Nuclear Physics.

The meeting is divided in four sessions :

1. General information
   • comprehensive overview, possibilities, issues and technologies to be presented
2. Practical exemple(s) 
   • goal: show one (or more) concrete application(s) 
3. Feedbacks / expectations from the community
   • goal: to gather needs (actions), to provide hints to start with ...
4. Table ronde : discussions 

 

*** Call for contributions closed ***

If you would like to give a talk, in particular in sessions 2 and 3, do not hesitate to send contributions (even if the work about Machine learning has not yet started) by giving a short explanation about the work and by providing a temporary title. 

Adrien, Eric et Olivier

 

mar. 29 octobre

10:00 → 11:00 Accueil

10:00 Café d'accueil

10:30 Introduction

IntroJourneeML.pdf

11:00 → 15:30 Introduction générale

11:00 Cours d'introduction au réseaux de neurones, avec description de quelques algorithmes populaires

Orateur: Julien Donini (LPC)

Julien.pdf

12:30 Déjeuner

14:00 Review des méthode de ML utilisées en physique de particules

Orateur: Yann Coadou (CPPM, Aix-Marseille Université, CNRS/IN2P3)

MLetPhysiqueNucleaire.pdf

15:30 → 20:30 Examples d'applications

15:30 Utilisation du machine learning en imagerie radio-isotopique per-opératoire

Orateur: Françoise Bouvet (IMNC)

ml-gammacamera-imnc-FB-oct19.pdf

16:15 Pause café

16:45 Machine Learning pour la discrimination gamma-neutron : études

Orateur: Xavier Fabian (IPN Lyon)

XFabian.pdf

17:15 Machine Learning pour la discrimination gamma-neutron : implementation

Orateur: Guillaume BAULIEU (IPNL)

Discrimination gamma/neutron : implémentation

17:45 Table ronde

18:30 Cocktail

mer. 30 octobre

09:00 → 13:00 Machine Learning en Physique Nucléaire

09:00 Intelligence artificielle et machine learning pour l'étude de réacteurs et de scénarios électro-nucléaires

Orateur: Marc ERNOULT (IPNO)

MLscenarionucleaires.pdf

09:30 Algorithme d'auto-apprentissage sur des signaux de courant issus de détecteurs silicium.

Orateur: Marin Chabot (Institut de physique Nucléaire d'Orsay)

GT5_chabot_30_10_2019.pdf

10:00 GPU, Machine Learning et Online

Orateur: Vincent LAFAGE (CNRS)

GDR_RESAnet_201910.2.pdf

10:30 Machine Learning @ ILL

Recently, by using deep learning methods, computers are able to surpass or come close to matching human performance on image analysis and pattern recognition. This advanced method could also help interpreting data from neutron scattering experiments. Those data contain rich scientific information about structure and dynamics of materials under investigation, and deep learning could help researchers better understand the link between experimental data and materials properties. We applied deep learning techniques to scientific neutron scattering data. This is a complex problem due to the multi-parameter space we have to deal with. We have used a convolutional neural network-based model to evaluate the quality of experimental neutron scattering images, which can be influenced by instrument configuration, sample and sample environment parameters. Sample structure can be deduced during data collection that can be therefore optimised. The neural network model can predict the experimental parameters to properly setup the instrument and derive the best measurement strategy. This results in a higher quality of data obtained in a shorter time, facilitating data analysis and interpretation.

Orateur: Paolo Mutti (ILL)

Orsay.pdf

11:00 Pause café

11:30 Tracking particles in an active target

Orateur: Thomas ROGER (GANIL)

ACTAR_MLPN_10-2019.pdf

11:45 learning to unmix in gamma-ray spectrometry

In the context of gamma-ray spectrum analysis, spectral unmixing tackles the activity estimation problem as an inverse problem, where individual activities appear as mixing weights related to individual spectra. Current approaches are agnostic to the available archive of past measurements, which bring valuable information to perform accurate radionuclide activity estimation. For that purpose, we propose unrolling methods to learn a data-driven prior from the available archive of measurements. Preliminary results show the efficiency of the proposed approach, with improved estimation accuracy both in estimation bias and variance.

Orateur: Jiaxin Xu (LMRE / IRSN)

pre_30oct.pdf

12:00 Taming nuclear complexity using deep neural networks

Orateur: Raphael-David Lasseri (CEA)

12:30 NPB, MCMC, GPE and other funny acronyms

In my talk i will discuss parameter estimate of a simple liquid drop model using at first simple Monte Carlo methods. The goal is to capture the correlations in the residuals and get a better estimate of error bars.
I will then apply Gaussian Process Emulator to investigate how to study the chi2 surface and speed up the convergence process.

Orateur: Alessandro Pastore (University of York)

Pastore.pdf