题 目： Approaches to Multistage One-Shot Decision Making

报告人： 日本横滨国立大学郭沛俊教授

时 间： 2014年09月24日 下午 1:30

地 点： 数学楼629教研室

摘要：Multistage decisions refer to a series of interdependent decisions which are made at each stage to achieve a final result. Dynamic programming is a powerful vehicle for dealing with multistage decision-making problems. The stochastic dynamic programming problems are used for handling multistage decision making under risk and are formulated as a maximization problem of the expected value with the assumption that the system under control is a Markov chain.

One-shot decision theory has been proposed by Guo for dealing with the decision problems under uncertainty in which the decision maker has one and only one chance for making a decision. In this research, we apply the one-shot decision theory to multistage one-shot decision problems under uncertainty. Multistage one-shot decision making is typical for such decision problems where at each stage a decision maker has one and only one chance to make a decision. The optimality equation in multistage one-shot decision problems is given. Comparing with the optimality equation in stochastic dynamic programming problems, the payoff associated with some specified scenario in multistage one-shot decision problems takes the place of the expected payoff in stochastic dynamic programming problems. The sequence of optimal decisions is determined by the proposed backward induction. We analyze the one-shot optimal stopping problem in Markov processes. The one-step look-ahead rule in the one-shot optimal stopping problem is proposed and other analytic results are obtained.