torchquad: Numerical Integration in Arbitrary Dimensions with PyTorch
Pablo Gómez, Håvard Hem Toftevaag, Gabriele Meoni
Abstract
torchquad is a Python module for n-dimensional numerical integration optimized for graphics processing units (GPUs). Various deterministic and stochastic integration methods, such as Newton-Cotes formulas and Monte Carlo integration methods like VEGAS Enhanced (Lepage, 2020), are available for computationally efficient integration for arbitrary dimensionality n d . As it is implemented using PyTorch (Paszke et al., 2019), one of the most popular machine learning frameworks, torchquad provides fully automatic differentiation throughout the integration, which is essential for many machine learning applications.
Topics & Concepts
Computer scienceComputational scienceParallel computingAdvanced Numerical Methods in Computational MathematicsMatrix Theory and AlgorithmsNumerical Methods and Algorithms