16-21 July 2017
Embassy Suites Buffalo
US/Eastern timezone

Generative Adversarial Networks for Jet Simulation

18 Jul 2017, 12:00
20m
Embassy Suites Buffalo

Embassy Suites Buffalo

200 Delaware Avenue Buffalo, NY 14202

Speakers

Michela Paganini (Yale University (US)) Luke Percival De Oliveira Ben Nachman (Lawrence Berkeley National Lab. (US))

Description

We introduce the first use of deep neural network-based generative modeling for high energy physics (HEP). Our novel Generative Adversarial Network (GAN) architecture is able cope with the key challenges in HEP images, including sparsity and a large dynamic range. For example, our Location-Aware Generative Adversarial Network learns to produce realistic radiation patterns inside high energy jets simultaneously for boosted W boson and generic quark and gluon jets. The pixel intensities of the GAN-generated images faithfully span many orders of magnitude and reproduce the distributions of important low-dimensional physical properties (e.g. jet mass, n-subjettiness, etc.). We provide many visualizations of what the GAN has learned, to build additional confidence in the algorithm. Our results demonstrate that high-fidelity, fast simulation through GANs is a promising application of deep neural networks for solving one of the most important challenges facing HEP, and in particular jet-substructure, today.

Primary authors

Michela Paganini (Yale University (US)) Luke Percival De Oliveira Ben Nachman (Lawrence Berkeley National Lab. (US))

Presentation Materials