Continuously Tempered PDMP samplers

Authors: Matthew Sutton, Robert Salomone, Augustin Chevallier, Paul Fearnhead

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show empirically that it can outperform existing PDMP-based samplers on challenging multimodal posteriors.
Researcher Affiliation Academia Matthew Sutton Centre for Data Science Queensland University of Technology matt.sutton@qut.edu.au Robert Salomone Centre for Data Science Queensland University of Technology robert.salomone@qut.edu.au Augustin Chevallier Department of Mathematics and Statistics Lancaster University a.chevallier@lancaster.ac.uk Paul Fearnhead Department of Mathematics and Statistics Lancaster University p.fearnhead@lancaster.ac.uk
Pseudocode Yes Algorithm 1 Zig-Zag algorithm
Open Source Code Yes Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] See Supplementary Material.
Open Datasets No The paper uses synthetic or conceptual problem setups like "Mixture of Gaussians", "Transdimensional example", and "Boltzmann machine relaxation", but does not provide specific access information (URL, DOI, or a formal citation with author/year) for a publicly available or open dataset used in these examples. While some are well-known problem types, no specific dataset access is given.
Dataset Splits No The paper describes using the first 40% of simulated event-times as burn-in, but this is not equivalent to a training/validation/test dataset split for model evaluation.
Hardware Specification Yes Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [Yes] See Supplementary Material.
Software Dependencies No The paper does not list specific software dependencies with version numbers in the main text. Although the checklist mentions the supplementary material, the prompt requires explicit details in the paper or a direct reference to a versioned software component.
Experiment Setup Yes All methods were simulated for 50,000 event-times, with the first 40% used as burnin in the standard Zig-Zag and used for both burn-in and estimating the polynomial κ(β) in the tempered samplers. For this example we choose q0(x) to be a Gaussian N(ν, Σ) centred at ν = (5, 5)T with covariance matrix Σ = 2I2.