IML Machine Learning Working Group

Tuesday 7 Jul 2020, 15:00 → 18:00 Europe/Zurich

40/S2-C01 - Salle Curie (CERN)

Meeting will be by video only on Zoom.

Join Zoom Meeting

https://cern.zoom.us/j/94718991390?pwd=VDc1dEtGNlNGU1Z2U1dMSWJkSGprZz09
 

Meeting ID: 947 1899 1390
Password:  check email

Recording: https://videos.cern.ch/record/2723303 (the recording currently works with Firefox, but there are problems with Safari, we are working to solve the issue and apologize for any inconvenience this may cause)

1 News

Speakers: Andrea Wulzer (CERN and EPFL), David Rousseau (LAL-Orsay, FR), Gian Michele Innocenti (CERN), Lorenzo Moneta (CERN), Loukas Gouskos (CERN), Dr Pietro Vischia (Universite Catholique de Louvain (UCL) (BE)), Riccardo Torre (CERN)

IML_Intro.pdf

2 Machine learning for lattice field theory

I will describe avenues to accelerate and enable lattice field theory calculations using machine learning. I will focus in particular on the role of generative models, and requirements such as guarantees of exactness in sampling and the incorporation of complex symmetries (e.g., gauge symmetry) into ML architectures.

Speaker: Phiala Shanahan (MIT)

Shanahan.pdf

3 Deep Learning as a Tool for Generic Searches at Colliders

In this talk, we will walk through current applications of Deep Learning as a tool for generic searches for New Physics at colliders. We will show recent results on how Deep Learning discriminators perform on new signal unseen during training and on unsupervised methods to search for New Physics in a complete signal agnostic approach.

Speaker: Dr Miguel Crispim Romao (LIP)

IML-WG-Jul-2020.pdf

4 Graph Neural Networks for 2D Calorimetric Cluster Reconstruction

In this work, we seek to explore the use of graph neural networks (GNNs) to perform cluster reconstruction from 2D readouts of a calorimeter system. We leverage the ability of GNNs to learn arbitrary detector geometries to regress in a multi-tasked fashion the energy and centroid coordinated of simulated deposits. This talk will focus on the preliminary results of a proof-of-concept study, amounting to training the algorithm on a simulated dataset loosely inspired by the LHCb electromagnetic calorimeter.

Speaker: Blaise Raheem Delaney (University of Cambridge (GB))

IML_Delaney_07072020.pdf