Multi-variable Integration with a Neural Network

Abstract

In this article we present a method for automatic integration of parametric integrals over the unit hypercube using a neural network. The method fits a neural network to the primitive of the integrand using a loss function designed to minimize the difference between multiple derivatives of the network and the function to be integrated. We apply this method to two example integrals resulting from the sector decomposition of a one-loop and two-loop scalar integrals. Our method can achieve per-mil and percent accuracy for these integrals over a range of invariant values. Once the neural network is fitted, the evaluation of the integral is between 40 and 125 times faster than the usual numerical integration method for our examples, and we expect the speed gain to increase with the complexity of the integrand.

15 pages, 4 figures, typo corrected, version accepted by JHEP

References in corpus (2)

• Machine learning materials physics: Integrable deep neural networks enable scale bridging by learning free energy functions
• Phase Space Sampling and Inference from Weighted Events with Autoregressive Flows

Cited by in corpus (7)

• MadNIS -- Neural Multi-Channel Importance Sampling
• Efficient phase-space generation for hadron collider event simulation
• Neural Control Variates with Automatic Integration
• Multi-variable integration with a variational quantum circuit
• Learning Feynman integrals from differential equations with neural networks
• Faster Bayesian inference with neural network bundles and new results for $f(R)$ models
• Variance Reduction via Simultaneous Importance Sampling and Control Variates Techniques Using Vegas