SQUIDS Seminar: Bayesian Experimental Design and Entropy Computation
| Dates: | 17 September 2025 |
|---|
| Times: | 11:00 - 12:00 |
|---|
| What is it: | Seminar |
|---|
| Organiser: | Department of Mathematics |
|---|
| Who is it for: | University staff, Current University students |
|---|
| Speaker: | Jinglai Li |
|---|
|
|
Speaker: Jinglai Li (Birmingham)
Abstract: Bayesian experimental design provides a principled approach for selecting experiments that maximize expected information gain. A central challenge in this framework is the computation of entropy and related information-theoretic quantities, which are essential for defining and optimizing design criteria. Such computation involves both estimation—evaluating entropy and Kullback–Leibler divergence in complex or likelihood-free models—and optimization, which requires efficient evaluation of entropy gradients.
In this talk, I will present some of our recent works addressing these challenges. First, I will introduce a normalizing-flow method for scalable entropy estimation in high-dimensional settings. Second, I will describe an approximate KLD-based design criterion that enables experimental design when the likelihood is intractable. Finally, I will discuss new techniques for estimating entropy gradients, which make gradient-based optimization feasible in Bayesian design.
Speaker
Jinglai Li
Organisation: University of Birmingham
Travel and Contact Information
Find event
Frank Adams Room 1
Alan Turing Building
Manchester
