HOGWILD!-Gibbs can be PanAccurate

Authors: Constantinos Daskalakis, Nishanth Dikkala, Siddhartha Jayanti

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

Reproducibility Variable Result LLM Response
Research Type Experimental We perform experiments on a multiprocessor machine to empirically illustrate our theoretical findings.
Researcher Affiliation Academia Constantinos Daskalakis EECS & CSAIL, MIT costis@csail.mit.edu Nishanth Dikkala EECS & CSAIL, MIT nishanthd@csail.mit.edu Siddhartha Jayanti EECS & CSAIL, MIT jayanti@mit.edu
Pseudocode Yes Input: Set of variables V , Configuration x0 S|V |, Distribution π initialization; for t = 1 to T do Sample i uniformly from {1, 2, . . . , n}; Sample Xi Prπ [.|X i = x i] and set xi,t = Xi; For all j = i, set xj,t = xj,t 1; end Algorithm 1: Gibbs Sampling
Open Source Code No The paper does not provide any statement or link indicating that the source code for their methodology is open-source or publicly available.
Open Datasets No The paper describes using Curie-Weiss and Grid models and generating samples by running Markov chains, but it does not provide concrete access information (link, DOI, formal citation) for a specific publicly available dataset used for the experiments.
Dataset Splits No The paper does not describe specific training, validation, or test dataset splits (percentages, sample counts, or predefined splits) for reproducibility. It discusses generating samples and comparing two methods directly.
Hardware Specification Yes We show the results of experiments run on a machine with four 10-core Intel Xeon E7-4850 CPUs
Software Dependencies No The paper does not provide specific software names with version numbers (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes To generate samples, we start at a uniformly random configuration and run Markov chains for T = 10n log2(n) steps to ensure mixing. We chose α = 0.5 to ensure Dobrushin s condition. Four asynchronous processors were used to generate the first plot, while twenty were used for the second. Each red point is the empirical mean of the function f computed over 5000 samples from the HOGWILD! Markov chain corresponding to CW(n, 0.5), and each blue point is the empirical mean produced from 5000 sequential runs of the same chain.