Ultra-Fast Event Generation using Modern Neural Networks

在当代数据科学（包括现代机器学习（ML）技术）的推动下，粒子物理学正在经历快速转型。这些方法的应用是由基础物理问题与快速第一原理模拟，庞大的数据集和完全不确定性控制的独特组合驱动的。这一发展的重点和目标是基于机器学习的仿真和分析工具，将事件生成器与基于仿真的推理方法相结合。我们的目标是建立第一个基于ML的矩阵元素事件生成器，开启高能物理中超快速事件生成的时代，在现象学，数据分析和解释方面具有广泛的应用前景。

Particle physics is undergoing a rapid transformation driven by contemporary data science, including modern machine learning (ML) techniques. Applications of such methods are driven by a unique combination of fundamental physics questions with fast first-principle simulations, vast datasets, and full uncertainty control. The keystone in this development and the goal of this proposal are ML-based simulation and analysis tools, combining event generators with simulation-based inference methods. Our ambition is to build the first ML- based matrix-element event generator, opening the era of ultra-fast event generation in high-energy physics, with a vast range of possible applications in phenomenology, data analysis, and interpretation.