Phase space sampling and inference from weighted events with autoregressive flows

Abstract

We explore the use of autoregressive flows, a type of generative model with tractable likelihood, as a means of efficient generation of physical particle collider events. The usual maximum likelihood loss function is supplemented by an event weight, allowing for inference from event samples with variable, and even negative event weights. To illustrate the efficacy of the model, we perform experiments with leading-order top pair production events at an electron collider with importance sampling weights, and with next-to-leading-order top pair production events at the LHC that involve negative weights.

A factorisation-aware Matrix element emulator

“…We investigated 50 different random trajectories. 4 For the discussion in this article we chose one that contains many interesting features, namely the matrix elements span many JHEP11(2021)066 Absolute percentage difference Absolute percentage difference orders of magnitudes and there are distinct peaks in the trajectory. In figure 7 we show the predictions by the three 5-jet models trained on data with different global phase-space cuts for this trajectory.…”

“…Since then neural networks and other machine learning techniques have proved useful in many other areas of the field. On the theory prediction side they have been used to improve the efficiency of Monte Carlo sampling [1][2][3][4][5], to accelerate the simulation of radiation within a jet [6][7][8], to streamline the processes of generation and unweighting of simulated event samples [9][10][11][12][13][14][15][16][17][18]. Closer to the experimental measurements they have also been used to emulate detector simulation [19][20][21][22], they can be used to perform unfolding [23] or correcting for detector effects [24], and perform pileup subtraction [23].…”

How to GAN Event Unweighting

“…While we were finalizing this study, similarly promising ideas were presented in Ref. [61], showing how generative networks benefit from training on weighted events.…”

Optimising simulations for diphoton production at hadron colliders using amplitude neural networks

“…There has also been a large focus on using ML for other components of MC event generator simulations. Specifically, Generative Adversarial Networks (GANs) [35] are being applied to event generation [36][37][38][39][40][41][42][43][44][45][46][47], event unweighting [48,49] and subtraction [50], with recent works incorporating Bayesian methods for uncertainty estimation into these generative methods [51]. NN-based approaches (some of which also use GAN technology) applied to parton showering [52][53][54][55] and event reweighting [56] have also been developed.…”