- This event has passed.
AMS Weekly Seminar | Raul Astudillo
February 6 @ 1:30 pm - 2:30 pm
Location: Gilman 50
When: February 6th at 1:30 p.m.
Title: Beyond the Black Box: Structure-Aware Bayesian Optimization for Efficient and Scalable Experimentation
Abstract: Experimentation is fundamental to scientific progress. However, the complexity of modern experimentation tasks surpasses the capabilities of trial-and-error approaches guided by human intuition. AI-driven frameworks for adaptive experimentation have emerged as powerful alternatives. Among them, Bayesian optimization—a machine learning framework for optimizing expensive-to-evaluate objective functions—has become one of the most successful, with applications ranging from hyperparameter tuning of deep neural networks to robot control.
Traditionally, Bayesian optimization methods have been employed as black-box optimizers. While this approach has achieved significant success, it often falls short in complex applications such as materials design and drug discovery. In this talk, I will discuss how exploiting composite and network-like structures inherent to many experimentation tasks can dramatically improve the efficiency and scalability of Bayesian optimization methods, enabling breakthroughs across various fields. I will conclude by discussing several exciting directions for future work in the broader field of AI-driven decision-making and discovery.
Zoom link: https://wse.zoom.us/j/93287142219?pwd=z9fqWnRMzmzS0SGijRiie5yN3kHRSZ.1