講    題：Dynamic Inventory Optimization with Learning and Model Ambiguity

主 講 人： 莊雅棠Chuang,Ya-Tang 教授

(國立成功大學工業與資訊管理學系)

主 持 人： 陳文智  教授

主辦單位：交通大學工業工程與管理系

時間：110年03月08日(星期一) 13:20~15:10

地    點：管二館520室

演講摘要：

This research concerns optimal inventory control in the presence of model ambiguity and statistical learning. Specifically, decision makers, on the one hand, face inventory control problems where parameters of the demand distribution are not known a priori, and need to be learned using right-censored sales data. On the other hand, the decision makers fear that the underlying model may be misspecified and attempt to find a robust policy against model ambiguity. Inventory control problems with learning are usually modeled using Bayesian dynamic programming (BDP). Under the Bayesian paradigm, it is assumed that the family of the demand distribution is known and data is generated from the presumed demand distribution. This assumption is however, not always satisfied for most practical problems. Another modeling approach, robust optimization, is usually used when the model is misspecified, and the decision makers hence solve a worst-case objective to hedge against model ambiguity. In this approach, data is essentially of no value when it comes to learning the model. Motivated by this concern, the goal of this research is to (i) develop new modeling frameworks that allow the decision makers to remain robust with respect to model ambiguity while also learning at the same time, and (ii) understand the effects of learning and robustness on the optimal objective values and decisions of the proposed model. The primary focus of our analysis is on establishing structural results of the decision makers’ optimal decisions. Our main result shows that the optimal decision can be expressed as the sum of a myopic decision plus a (non-negative) learning boost, minus a (non-negative) robust adjustment and an (non-negative) interaction adjustment. Moreover, through this representation, we find that the famous “stock more” result could be sometimes revised to stock less even if the benefit from learning is increasing.

演講性質： 學術研究專題

歡迎聽講