Discussion of performance enhancement of Hoeffding probability inequality for average of data from different populations

Ikuo Arizono, Ryousuke Tomohiro, Takanori Asahara, Yasuhiko Takemoto

Research output: Contribution to journalArticle

Abstract

Probability inequalities are used as a means to evaluate the upper bounds of upper probability for an average of some random variables based on finite stochastic properties, excluding the probability distribution functions of respective random variables. Note that in the case of applying the probability inequality, it is not necessary for the population of each random variable to be identical. Of the existing probability inequalities, Hoeffding probability inequality is highly regarded as an excellent probability inequality from the viewpoint of performance. Hoeffding probability inequality evaluates the upper bounds of upper probability for an average of some random variables based on limited stochastic properties such as expectation, variance and codomain in each random variable. However, we consider that there is a room to improve the performance of Hoeffding probability inequality. In this study, we demonstrate a technique for improving the performance of Hoeffding probability inequality in the case that the population of each random variable is not always identical.

Original languageEnglish
Pages (from-to)61-69
Number of pages9
JournalJournal of Japan Industrial Management Association
Volume65
Issue number2
Publication statusPublished - Jan 1 2014

Keywords

  • Hoeffding probability inequality
  • Upper bound
  • Upper probability

ASJC Scopus subject areas

  • Strategy and Management
  • Management Science and Operations Research
  • Industrial and Manufacturing Engineering
  • Applied Mathematics

Fingerprint Dive into the research topics of 'Discussion of performance enhancement of Hoeffding probability inequality for average of data from different populations'. Together they form a unique fingerprint.

  • Cite this