Ensuring Rapid Mixing and Low Bias for Asynchronous Gibbs Sampling

Authors: Christopher De Sa, Chris Re, Kunle Olukotun

ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show experimentally that our theoretical results match practical outcomes. and We validate our results experimentally and show that, by using asynchronous execution, we can achieve wall-clock speedups of up to 2.8 on real problems.
Researcher Affiliation Academia Christopher De Sa CDESA@STANFORD.EDU Department of Electrical Engineering, Stanford University, Stanford, CA 94309 Kunle Olukotun KUNLE@STANFORD.EDU Department of Electrical Engineering, Stanford University, Stanford, CA 94309 Christopher R e CHRISMRE@STANFORD.EDU Department of Computer Science, Stanford University, Stanford, CA 94309
Pseudocode Yes Algorithm 1 Gibbs sampling Require: Variables xi for 1 i n, and distribution π. for t = 1 to T do Sample s uniformly from {1, . . . , n}. Re-sample xs uniformly from Pπ(Xs|X{1,...,n}\{s}). end for
Open Source Code No The paper does not provide concrete access to source code for the methodology described, such as a specific repository link or an explicit statement of code release.
Open Datasets Yes we ran HOGWILD!-Gibbs sampling on a real-world 11 GB Knowledge Base Population dataset (derived from the TAC-KBP challenge)
Dataset Splits No The paper describes the datasets used (e.g., synthetic Ising model graph, KBP dataset) but does not provide specific details on how these datasets were split into training, validation, or test sets.
Hardware Specification Yes using a machine with a single-socket, 18-core Xeon E7-8890 CPU and 1 TB RAM.
Software Dependencies No The paper does not specify any software dependencies with version numbers.
Experiment Setup Yes we simulated HOGWILD!-Gibbs sampling running on a random synthetic Ising model graph of order n = 1000, degree = 3, inverse temperature β = 0.2, and prior weights Ex = 0.