Learning to simulate high energy particle collisions from unlabeled data

• DOI: 10.1038/s41598-022-10966-7
• Publication Date: 2022-05-09T13:04:52Z
• Authors: Jessica N. Howard; Stephan Mandt; Daniel Whiteson; Yibo Yang
• Keywords: Autoencoder; Experimental data

In many scientific fields which rely on statistical inference, simulations are often used to map from theoretical models experimental data, allowing scientists test model predictions against results. Experimental data is reconstructed indirect measurements causing the aggregate transformation be poorly-described analytically. Instead, numerical at great computational cost. We introduce Optimal-Transport-based Unfolding and Simulation (OTUS), a fast simulator based unsupervised machine-learning that capable of predicting models. Without aid current simulation information, OTUS trains probabilistic autoencoder transform directly between data. Identifying autoencoder's latent space with causes decoder network become fast, predictive potential replace current, computationally-costly simulators. Here, we provide proof-of-principle results two particle physics examples, Z-boson top-quark decays, but stress can widely applied other fields.