Random Feature Stein Discrepancies

Authors: Jonathan Huggins, Lester Mackey

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In our experiments with sampler selection for approximate posterior inference and goodness-of-fit testing, RΦSDs perform as well or better than quadratic-time KSDs while being orders of magnitude faster to compute.
Researcher Affiliation Collaboration Jonathan H. Huggins Department of Biostatistics, Harvard Lester Mackey Microsoft Research New England
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes See https://bitbucket.org/jhhuggins/random-feature-stein-discrepancies for our code.
Open Datasets No The paper uses standard datasets like the bimodal Gaussian mixture model and multivariate Gaussian/Laplace/t-distributions, which are widely known, but it does not provide explicit access information (link, DOI, specific citation with author/year) for these datasets in the text. It refers to a previous paper for the GMM target but does not explicitly link or cite it here for the dataset.
Dataset Splits No The paper does not explicitly provide details about training, validation, and test splits for the datasets used in the experiments. It mentions using a small subsample of the full dataset for certain calculations, but not explicit splits for model training/evaluation.
Hardware Specification No The paper does not explicitly describe the hardware used for running its experiments.
Software Dependencies No The paper does not provide specific version numbers for software dependencies used in the experiments.
Experiment Setup Yes In particular, we chose = γ/3, λ = 1 /2, and = 4 /(2 + ). Except for the importance sample efficiency experiments, where we varied γ explicitly, all experiments used γ = 1/4. Let d medu denote the estimated median of the distance between data points under the u-norm, where the estimate is based on using a small subsample of the full dataset. For L2 Sech Exp, we took a 1 = 2 d med1, except in the sample quality experiments where we set a 1 = 2 . For most experiments we took β = 1/2, c = 4 d med2, and df = 0.5. The exceptions were in the sample quality experiments, where we set c = 1, and the restricted Boltzmann machine testing experiment, where we set c = 10 d med2 and df = 2.5.