On the benefits of Laplace samples in solving a rare event problem using cross-entropy method
作者：
Highlights：
• Influence of long-tailed distributions in the cross-entropy optimizer, on generating candidate solutions is analyzed.
• A theoretical reasoning for the superior performance of Laplace samples has been provided.
• A rigorous empirical study has been conducted to verify our conjecture.
• In exploration and exploitation, Laplace samples strike a balance between the Gaussian and Cauchy/logistic samples.
• The findings can be extended to other heuristic/evolutionary algorithms meant for rare event problem solving.
摘要
•Influence of long-tailed distributions in the cross-entropy optimizer, on generating candidate solutions is analyzed.•A theoretical reasoning for the superior performance of Laplace samples has been provided.•A rigorous empirical study has been conducted to verify our conjecture.•In exploration and exploitation, Laplace samples strike a balance between the Gaussian and Cauchy/logistic samples.•The findings can be extended to other heuristic/evolutionary algorithms meant for rare event problem solving.
论文关键词：Cross-entropy optimization,Long-tailed distributions,Nonconvex fitness function
论文评审过程：Available online 12 November 2013.
论文官网地址：https://doi.org/10.1016/j.amc.2013.10.011
