CMX Student/Postdoc Seminar

Experimentation is a cornerstone of science and a key driver of human progress. Many experimental tasks can be viewed as optimization problems, where the goal is to find the best solution under constraints of cost, time, or other resources. Bayesian optimization has emerged as a powerful method for solving these problems. However, tasks in high-stakes fields such as materials design and drug discovery often present challenges that surpass the capabilities of existing approaches.

In this talk, I will discuss recent advances that overcome these limitations. Specifically, I will discuss how leveraging the composite structure present in many experimental tasks can dramatically improve the efficiency and scalability of Bayesian optimization. This approach opens new avenues for tackling complex problems out of the reach of standard methods. Lastly, I will outline future research directions toward developing a comprehensive framework for efficient, adaptive experimental design.