Probabilistic Exponential Integrators

Authors: Nathanael Bosch, Philipp Hennig, Filip Tronarp

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the proposed methods on multiple stiff differential equations and demonstrate their improved stability and efficiency over established probabilistic solvers. (Abstract) and 4 Experiments In this section we investigate the utility and performance of the proposed probabilistic exponential integrators and compare it to standard non-exponential probabilistic solvers on multiple ODEs.
Researcher Affiliation Academia Nathanael Bosch Tübingen AI Center, University of Tübingen nathanael.bosch@uni-tuebingen.de Philipp Hennig Tübingen AI Center, University of Tübingen philipp.hennig@uni-tuebingen.de Filip Tronarp Lund University filip.tronarp@matstat.lu.se
Pseudocode No The paper does not contain a clearly labeled pseudocode or algorithm block.
Open Source Code Yes Code for the implementation and experiments is publicly available on Git Hub.1
Open Datasets No The paper defines the differential equations and their parameters directly within the text and appendix, rather than utilizing external, pre-existing public datasets that require explicit access information.
Dataset Splits No The paper solves differential equations numerically and does not describe data splitting into training, validation, or test sets in the typical machine learning sense.
Hardware Specification No All experiments run on a single, consumerlevel CPU. - This statement lacks specific details such as the CPU model, number of cores, or clock speed.
Software Dependencies No All methods are implemented in the Julia programming language [1]... Reference solutions are computed with the Differential Equations.jl package [34]. - The paper mentions the software used but does not provide specific version numbers for Julia or Differential Equations.jl.
Experiment Setup Yes We start with a simple one-dimensional initial value problem: a logistic model with negative growth rate parameter r = 1 and carrying capacity K R+, of the form y(t) = y(t) + 1/K y(t)2, t [0, 10], y(0) = 1. (Section 4.1) and We transform the problem into a semi-linear ODE with the method of lines [29, 38], and discretize the spatial domain on 250 equidistant points and approximate the differential operators with finite differences. (Section 4.2) and We again consider a domain Ω= (0, 1), which we discretize on a grid of N = 100 points. (Section 4.3)