Hamiltonian Monte Carlo Without Detailed Balance

Authors: Jascha Sohl-Dickstein, Mayur Mudigonda, Michael DeWeese

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate a greater than factor of two improvement in mixing time on three test problems. We release the source code as Python and MATLAB packages. As illustrated in Figure 2, we compare the mixing time for our technique and standard HMC on three distributions.
Researcher Affiliation Collaboration Jascha Sohl-Dickstein JASCHA@STANFORD.EDU Stanford University, Palo Alto. Khan Academy, Mountain View Mayur Mudigonda MUDIGONDA@BERKELEY.EDU Redwood Institute for Theoretical Neuroscience, University of California at Berkeley Michael R. De Weese DEWEESE@BERKELEY.EDU Redwood Institute for Theoretical Neuroscience, University of California at Berkeley
Pseudocode No The paper describes algorithms in numbered lists (Section 3.5 'Standard HMC' and Section 4.2 'Algorithm' for LAHMC) but does not use explicit 'Pseudocode' or 'Algorithm' labels as a block title or figure.
Open Source Code Yes We release the source code as Python and MATLAB packages. MATLAB and Python implementations of LAHMC are available at http://github.com/ Sohl-Dickstein/LAHMC.
Open Datasets No The paper describes the mathematical form of the distributions used for experiments (e.g., '2 and 100 dimensional ill-conditioned Gaussian distributions', 'isotropic quadratic and sinusoids') but does not provide specific links, DOIs, repository names, or formal citations for public access to these 'datasets'.
Dataset Splits No The paper evaluates sampling algorithms based on mixing time and autocorrelation but does not specify explicit training, validation, or test dataset splits (e.g., '80/10/10 split', sample counts, or cross-validation details) for reproducibility.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory, or cloud instance types) used for running the experiments.
Software Dependencies No The paper mentions that the source code is released as 'Python and MATLAB packages' but does not specify concrete version numbers for these or any other ancillary software libraries or dependencies.
Experiment Setup Yes HMC and LAHMC both had step length and number of leapfrog steps set to ϵ = 1, and M = 10. Values of β were set to 1 or 0.1 as stated in the legend. For LAHMC the maximum number of leapfrog applications was set to K = 4.