Distributed Bayesian Posterior Sampling via Moment Sharing
Authors: Minjie Xu, Balaji Lakshminarayanan, Yee Whye Teh, Jun Zhu, Bo Zhang
NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the speed and inference quality of our method with empirical studies on Bayesian logistic regression and sparse linear regression with a spike-and-slab prior. |
| Researcher Affiliation | Academia | 1State Key Lab of Intelligent Technology and Systems; Tsinghua National TNList Lab 1Department of Computer Science and Technology, Tsinghua University, Beijing 100084, China 2Gatsby Unit, University College London, 17 Queen Square, London WC1N 3AR, UK 3Department of Statistics, University of Oxford, 1 South Parks Road, Oxford OX1 3TG, UK |
| Pseudocode | No | The paper describes the algorithm in text but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper thanks other researchers for sharing their implementations but does not provide any statement or link for the open-sourcing of their own code. |
| Open Datasets | Yes | We experimented using the Boston housing dataset which consists of N = 455 training data points in d = 13 dimensions. |
| Dataset Splits | No | The paper mentions using simulated data and the Boston housing dataset but does not specify training, validation, or test splits, nor does it describe cross-validation. |
| Hardware Specification | Yes | Experiments were conducted on a cluster with as many as 24 nodes (Matlab workers), arranged in 4 servers, each being a multi-core server with 2 Intel(R) Xeon(R) E5645 CPUs (6 cores, 12 threads). |
| Software Dependencies | No | The paper states that methods were implemented and tested in Matlab, but it does not provide specific version numbers for Matlab or any other software dependencies. |
| Experiment Setup | Yes | The damping factor used was 0.2. At each EP iteration, SMS produced both the EP approximated Gaussian posterior q(θ; η0 +Pm i=1 ηi), as well as a collection of m T local posterior samples Θ. We use K to denote the total number of EP iterations. The sampler was initialised at 0d and used the first 20d samples for burn-in, then thinned every other sample. |