Accelerating Event Generation in Strong-Field QED with Neural Importance Sampling

Efficient Monte Carlo integreation is crucial for modeling processes at the European XFEL. However, traditional approaches to importance sampling like VEGAS do not perform well when integrands display multiple features or non-coordinate aligned features. In this work, we present an implementation of neural importance sampling (NIS) in the Julia programming language to address this challenge. We demonstrate the effectiveness of NIS by applying it to processes in strong-field QED at high energies, showing superior adaption of the integrand and thus enabling efficient event generation.