BriefGPT.xyz
Nov, 2013
渐近精确，尴尬地并行的 MCMC
Asymptotically Exact, Embarrassingly Parallel MCMC
HTML
PDF
Willie Neiswanger, Chong Wang, Eric Xing
TL;DR
该论文提出了一种并行的马尔可夫链蒙特卡罗算法，能够降低学习过程中的同步需求所带来的通信成本，并成功地实现了在多台机器上独立处理多个数据子集，从而生成大数据集的后验分布样本。
Abstract
Communication costs, resulting from
synchronization requirements
during learning, can greatly slow down many
parallel machine learning
algorithms. In this paper, we present a parallel
→