Download presentation
Presentation is loading. Please wait.
Published byArron Bryant Modified over 6 years ago
1
Machine Learning applied to CTA Thomas Vuillaume LAPP, September 2017
03/09/2018 Machine Learning applied to CTA Thomas Vuillaume LAPP, September 2017 Entrez votre nom
2
Standard machine learning
03/09/2018 Standard machine learning Standard machine learning Already in use in other IACT (e.g. boosted decision trees, random forest…) Use physical parameters derived from images (signal amplitude, ellipsoid length/width… ) Supervised learning Adaptation and developments for CTA Use of Python = standardisation with reknown packages (e.g. scikit-learn) Integration in LAPP pipeline and ctapipe Energy reconstruction Particles discrimination LAPP, 19/09/2017 Thomas Vuillaume Entrez votre nom
3
Standard machine learning
03/09/2018 Standard machine learning User friendly: LAPP, 19/09/2017 Thomas Vuillaume Entrez votre nom
4
Standard machine learning
03/09/2018 Standard machine learning Python makes exploration easy LAPP, 19/09/2017 Thomas Vuillaume Entrez votre nom
5
Standard machine learning
03/09/2018 Standard machine learning Developments done jointly with ctapipe’s Comparison and merging of latest results in one week during ctapipe developers workshop @LAPP September 25-29 LAPP, 19/09/2017 Thomas Vuillaume Entrez votre nom
6
GammaLearn project GammaLearn project Collaboration of three parties
03/09/2018 GammaLearn project GammaLearn project Collaboration of three parties Made possible under ASTERICS project LAPP, 19/09/2017 Thomas Vuillaume Entrez votre nom
7
03/09/2018 GammaLearn project ASTERICS call for expression of interest from industrial partners in December 2016 at 1st workshop Answer with Orobix, an italian start-up specializing in deep learning (life sciences, manufacturers…) LAPP, 19/09/2017 Thomas Vuillaume Entrez votre nom
8
18 months collaboration (until the end of ASTERICS)
03/09/2018 GammaLearn project 18 months collaboration (until the end of ASTERICS) Goal : CTA on-site particle discrimination Why? CTA ~5Gbps Data reduction 1 photon for > 100 proton = Most of signal is noise Data transfer requires a volume reduction of 1/10 Improve sensitivity Integration with LAPP pipeline and HPC softwares to make the most out of hybrid architectures LAPP, 19/09/2017 Thomas Vuillaume Entrez votre nom
9
03/09/2018 GammaLearn project LISTIC = Laboratoire d’Informatique, Traitement de l’Information et de la Connaissace with a team specializing in deep learning Collaboration proposal in the form of a co-directed PhD with two sources of fundings ASTERICS Fondation de l’Université Savoie Mont-Blanc with project accepted in June LAPP, 19/09/2017 Thomas Vuillaume Entrez votre nom
10
Different but complementary approach with Orobix
03/09/2018 GammaLearn project Collaboration with Orobix on tools and work environment = learning from industry Using Generative Adversarial Networks to reconstruct electromagnetic showers in 3D based on their 2D images from telescopes Different but complementary approach with Orobix LAPP, 19/09/2017 Thomas Vuillaume Entrez votre nom
11
MoU signed early September!
03/09/2018 GammaLearn project MoU signed early September! Kick-off meeting ! MUST, get prepared ! LAPP, 19/09/2017 Thomas Vuillaume Entrez votre nom
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.