MonteCarloMeasurements.jl: Nonlinear Propagation of Arbitrary Multivariate Distributions by means of Method Overloading

TL;DR

MonteCarloMeasurements.jl is a Julia package that enables easy propagation of multivariate probability distributions through arbitrary functions using Monte Carlo methods, with a user-friendly design that leverages method overloading.

Contribution

The package introduces a type-based approach for representing probability distributions with unweighted samples, simplifying nonlinear propagation and optimizing computational performance.

Findings

Efficient propagation of multivariate distributions through arbitrary functions.

Facilitates uncertainty propagation in ODE solvers.

Enables robust probabilistic optimization with automatic differentiation.

Abstract

This manuscript outlines a software package that facilitates working with probability distributions by means of Monte-Carlo methods, in a way that allows for propagation of multivariate probability distributions through arbitrary functions. We provide a \emph{type} that represents probability distributions by an internal vector of unweighted samples, \texttt{Particles}, which is a subtype of a \texttt{Real} number and behaves just like a regular real number in calculations by means of method overloading. This makes the software easy to work with and presents minimal friction for the user. We highlight how this design facilitates optimal usage of SIMD instructions and showcase the package for uncertainty propagation through an off-the-shelf ODE solver, as well as for robust probabilistic optimization with automatic differentiation.