Abstract

Likelihood free generative modeling for high energy physics

Benjamin Nachman
Lawrence Berkeley National Laboratory

I will discuss two techniques for generating new examples in high energy physics without having an explicit description of the probability density. One tool is the generative adversarial network (GAN) which is being used to accelerate simulation and can also be used for data-driven Standard Model background estimation. The second tool uses event reweighting to morph an existing set of events into another where the morphing function is a deep neural network. This tool can be further used to perform parameter estimation when combined with a second, fitting network.