Boosting high energy physics with generative networks

Anja Butter, Heidelberg University

upcoming: February 24, 2021

Abstract

Physicists at the Large Hadron Collider (LHC) are searching for signs of new physics to answer fundamental questions like the nature of dark matter. In many scenarios these signals are expected to be as rare as 1 in 10^10 events. We therefore require simulations which can model complex event structures with high precision. LHC physics is unique in the sense that we can rely on first-principles based predictions, which means that simulations rely on a small number of fundamental parameters to simulate observables over many orders of magnitude. I will show how generative neural networks can be used to supplement these simulations in order to match precision requirements of future collider experiments. In addition one can use flow based invertible networks to invert the simulations chain and unfold detector level events to understand the mechanisms at the heart of proton collisions.

Bio

Anja Butter is a postdoc at the ITP in Heidelberg, working on particle physics. Her research interests include Higgs physics, any signs of physics beyond the Standard Model, and the development of machine learning techniques to learn more about both. As the LHC is collecting large amounts of data in the search for new physics, machine learning has become an exciting technique to analyze and learn from these data. While it is crucial to push the limits of particle physics with new developments and techniques, the obtained results have to be understood in a global context. Therefore she is interested as well in global analyses of models like effective field theory and supersymmetry in which we can combine different measurements.

Event Info

The event will take place on Wednesday, February 24, 2021 at 11:00am . It will be a joint event with the DKFZ Data Science Seminar and will take place online. You can join the live stream at this URL! Kindly help us plan ahead by registering for the event on our meetup page.