“Optimal Sequential Decision Making” Seminar by Prof. Calvin Tsay (Imperial College London)

Many problems in chemical engineering require making good sequential decisions under uncertainty. In process operation, plants must be scheduled flexibly in response to time-varying electricity prices; in research, one must decide which costly experiments to run next. To explore solutions to these problems, we looked into the latest research through the seminar “Optimal Sequential Decision Making: Mixed-Integer Programming and Statistical Learning” by Prof. Calvin Tsay of the Department of Computing at Imperial College London, who has led research on decision-making frameworks that combine mathematical optimization and machine learning. The seminar covered methodologies in which statistical learning is used to build a surrogate model of the system, and mixed-integer programming (MIP) leverages that model to compute the next decision with guarantees on constraint satisfaction and optimality. Applications included safe, sample-efficient reinforcement learning for supply chain management and demand-response operation, as well as constrained and non-myopic Bayesian optimization for design of experiments, highlighting a central trade-off: investing more computation per decision can yield safer, better, and more far-sighted choices. The talk was followed by a lively Q&A session on topics such as how constraints are handled in different reinforcement learning methods and multi-step lookahead in Bayesian optimization.