Generative adversarial networks for physics research

Titel: Generative adversarial networks for physics research
Föreläsare: Jonas Glombitza, doktorand, RWTH Aachen, Tyskland

Abstract

Deep learning describes the state-of-the-art technique of machine learning with deep neural networks, i.e. neural networks with many layers. Whether for internet-search results, machine translation, image classification or speech recognition on the smartphone, deep learning is already part of our everyday life.

Recently, great progress has been made in the field of deep generative models using generative adversarial networks (GANs) and associated techniques. These developments offer various applicationsin physics research. Beside the cost-effective generation of simulation data, adversarial frameworks canbe used to refine simulated detector data or to provide powerful discrimination variables with reduced systematic uncertainties.

In this talk an introduction to adversarial frameworks and recent developments together with example applications in physics is given.

First, the concept of GANs and their challenging training is discussed. Subsequently, significant developments are presented which stabilize the training of GANs and improve their performance. Finally, example applications in physics are discussed ranging from the simulation of calorimeter data to the refinement of cosmic-ray simulations.