Deep learning with t-exponential Bayesian kitchen sinks

作者：

Highlights：

• We propose a novel Bayesian deep learning architecture.

• Our approach performs inference over the employed nonlinearities.

• Each layer comprises a bank of arbitrary samples from a nonlinearity.

• These are linearly combined using multiple alternative sets of weights.

• We train via approximate (variational) inference, based on a t-divergence measure.

摘要

•We propose a novel Bayesian deep learning architecture.•Our approach performs inference over the employed nonlinearities.•Each layer comprises a bank of arbitrary samples from a nonlinearity.•These are linearly combined using multiple alternative sets of weights.•We train via approximate (variational) inference, based on a t-divergence measure.

论文关键词：Random kitchen sinks,Student’s-t distribution,t-divergence,Variational bayes

论文评审过程：Received 23 November 2017, Revised 23 December 2017, Accepted 10 January 2018, Available online 12 January 2018, Version of Record 6 February 2018.

论文官网地址：https://doi.org/10.1016/j.eswa.2018.01.013