讲座题目:A Globally Optimal Actor-critic Framework for Solving Inventory Management Problems
报告人: 黄毅哲 德州大学奥斯汀分校
讲座时间:2023年5月26日(周五) 12:15-13:30
讲座地点:中国科学院大学中关村校区教学楼S406
腾讯会议 ID:753-627-934
内容摘要
Inventory management is one of the most important problems in logistics. Curses of dimensionality prevent us from finding optimal policies through naive dynamic programming in multi-stage settings with large state and action spaces. Previous studies have shown that under the vanilla model, the optimal policy is in a simple class of base stock policies, but this conclusion does not hold for more complex models with stochastic lead times and multi-echelon warehouses. We proposed an actor-critic framework with two neural networks that finds the optimal policy based on available data. It is shown that under several assumptions, the algorithm is guaranteed to find the globally optimal policy given enough data.
主讲人简介
黄毅哲,现为德州大学奥斯汀分校(University of Texas at Austin)商学院IROM三年级博士研究生。加州大学圣地亚哥分校(University of California San Diego)计算机系本科、硕士研究生毕业。