Speaker
Theo Heimel
(UCLouvain)
Description
The high-luminosity LHC will collect an unprecedented amount of data, and we will need to match it with fast and precise theory predictions. In recent years, ML-based sampling and surrogates have become key technologies for achieving these goals within the available computational budget. In this talk, I will review the ongoing efforts to establish fully GPU- and ML-accelerated sampling at leading order as a production-ready part of the MadGraph7 and Pepper event generators. I will also discuss how to extend these methods to NLO and NNLO to address challenges such as costly loop matrix-element evaluations and negative-weight events.