Behavior monitoring under uncertainty using Bayesian surprise and optimal action selection

作者：

Highlights：

• A probabilistic characterization is used as a specification for the desired behavior.

• A Bayesian surprise metric is defined for on-line monitoring of an action selection policy.

• Sparsification of data streams is used to reject outliers and leave out redundant information.

• Timely detection of performance degradation in an artificial pancreas is demonstrated.

摘要

•A probabilistic characterization is used as a specification for the desired behavior.•A Bayesian surprise metric is defined for on-line monitoring of an action selection policy.•Sparsification of data streams is used to reject outliers and leave out redundant information.•Timely detection of performance degradation in an artificial pancreas is demonstrated.

论文关键词：Artificial pancreas,Bayesian surprise,Behavior monitoring,Kullback–Leibler divergence,Optimal action selection

论文评审过程：Available online 6 May 2014.

论文官网地址：https://doi.org/10.1016/j.eswa.2014.04.031