Leigh Whitehead (Cambridge)

Model-Assisted Generative Adversarial Networks

Abstract: We propose and demonstrate the use of the model-assisted generative adversarial network (GAN) to perform fast production of images that accurately match the underlying simulation process. These images are produced in a physically motivated way by mimicking the simulation for a range of physics parameters describing the features seen in the images. In a second stage, the model-assisted GAN can extract the best-matching physics parameters from a set of true data by comparing the generated and true data images. We discuss two cases studies from the recent publication as well as a use case from the DUNE experiment. Finally, we conclude with some upcoming studies and extensions to the method.

Bio:
2008-2012 PhD on T2K at Warwick
2012-2016: Research associate on MINOS / MINOS+ / CHIPS at UCL
2016-2018: CERN research fellow working on DUNE and ProtoDUNE
2019-present: Senior research associate on DUNE and ProtoDUNE