Speaker: Jan-Matthis Lückmann , University of Tübingen
Science and industry make extensive use of simulations to model the world. However, conventional statistical inference is often inapplicable to detailed simulation models because their associated likelihood functions are intractable. In this talk, I discuss how simulation-based inference (SBI) addresses this problem, with an emphasis on applications to biophysical models of neural dynamics. I highlight SBI’s potential to close the gap between data-driven and theory- driven models in neuroscience. Furthermore, I present our recently-introduced, first-ever benchmark of SBI that compares algorithms in a transparent and reproducible way, explaining different approaches, metrics for principled comparisons, key findings, and open challenges.
Jan-Matthis Lückmann is currently finishing his PhD in Computer Science at the University of Tübingen, advised by Prof. Jakob Macke. Jan-Matthis’ research interests are at the intersection of machine learning and computational neuroscience. His expertise lies in simulation-based inference and its applications to mechanistic models. Jan-Matthis’ work includes the proposal of fast and flexible inference algorithms based on neural density estimation and their applications to biophysical models in neuroscience. Most recently, he introduced the first-ever benchmark for the rapidly developing field of simulation-based inference.
This meeting was not recorded.