Fast sampling methods for Bayesian max-margin models

作者：

Highlights：

• We study the stochastic subgradient MCMC methods for Bayesian max-margin models.

• Theoretical analysis shows the approximate detailed balance property of our methods.

• Stochastic subsampling and thermostats are used for fast convergence and mixing.

• Experiments show the efficiency and accuracy of our methods.

摘要

•We study the stochastic subgradient MCMC methods for Bayesian max-margin models.•Theoretical analysis shows the approximate detailed balance property of our methods.•Stochastic subsampling and thermostats are used for fast convergence and mixing.•Experiments show the efficiency and accuracy of our methods.

论文关键词：Inference,Stochastic MCMC,Subgradient MCMC,Bayesian max-margin models,Approximate detailed balance

论文评审过程：Received 27 March 2016, Revised 8 September 2016, Accepted 16 October 2016, Available online 17 October 2016, Version of Record 31 October 2016.

论文官网地址：https://doi.org/10.1016/j.eswa.2016.10.036