Abstract: Statistical inference of simulator-based models, used in many domains of science and engineering, is challenging due to the unavailability of the likelihood function, which is the probability density function of the data given the parameters. To solve this problem, the field of simulation-based inference (SBI) has emerged, wherein large numbers of simulations from the model are utilized for inference instead of the likelihood function. However, the performance of SBI methods severely degrades when the model fails to capture the real-world phenomenon under study (i.e., under model misspecification), or when the model is computationally costly to run, thus limiting the number of available simulations. In this talk, I will talk about some of my recent work on addressing these issues by developing robust and sample-efficient SBI methods.

Speaker: Ayush Bharti is a postdoctoral researcher at the Department of Computer Science, Aalto University, working with Prof. Samuel Kaski. He is affiliated with the Probabilistic Machine Learning research group at Aalto and the Finnish Center for Artificial Intelligence. He received the M.Sc. degree in signal processing and computing and the Ph.D. degree in wireless communications from Aalborg University, Denmark, in 2017 and 2021, respectively. Ayush’s research interests include simulation-based inference, Bayesian optimization, and Bayesian experimental design.