Approximate dynamic programming (ADP) and more broadly reinforcement learning (RL) have emerged as powerful tools to solve complex multi-stage optimization problems under uncertainty. Many of the game playing advances made by artificial intelligence approaches have at their heart some form of ADP or RL approach and they continue to be relevant to many areas of problem solving and optimization.
In the noughties, while Jay was at Georgia Tech, we pursued a very productive collaboration that applied these techniques to planning and scheduling problems of relevance to chemical engineering. In this talk I will review these approaches to multi-stage stochastic decision making and highlight the main contributions that we made during this time and through our continued collaboration after Jay left Georgia Tech.
