This seminar is organised as part of the Research Theme Scalable Approximation of Models and Algorithms (RT 3)
Title
Efficient Likelihood-free Monte Carlo Sampling for Generalized Bayesian Inference Using Probabilistic Surrogate Loss
Abstract
Likelihood-free inference (LFI) refers to techniques for inferring model parameters using simulations from complex models without relying on the analytic form of the simulator. In recent years, LFI has been significantly accelerated through the use of surrogate models that efficiently capture the relationship between parameters and observations, and leverage this information to select informative parameter values for querying the simulator. These surrogate-based LFI methods are often Bayesian, drawing posterior samples from an approximated posterior distribution. A flexible alternative to traditional Bayesian inference is the Generalized Bayesian Inference (GBI) framework, which relies on a highly general mechanism for updating parameter beliefs in light of observed data. The formulation of GBI makes it a natural complement to surrogate model–based LFI. We introduce a novel iterative algorithm that uses sequential Monte Carlo to simultaneously generate samples from the approximate GBI posterior and determine which parameter values should be queried to further improve the surrogate. This active learning strategy is guided by the uncertainty in Monte Carlo importance weights induced by the probabilistic surrogate models.
Bio
Assistant Professor Henri Pesonen, Faculty of Information Technology and Communication Sciences, Tampere University
Institutional website (external link)
Speaker is presenting in person.
Practical
The Tuesdays seminars series are devoted to various topics relevant for the Integreat´s research focus. Presenters from the Integreat community and beyond have 40 minutes to present, followed by a group discussion.
Seminars are open for attendance for everybody.
For those unable to attend in person: https://uio.zoom.us/j/62535744013
Point of contacts
- Organiser and moderator: Geir Olve Storvik, University of Oslo
- Integreat Administrative Leader Maria Dikova, University of Oslo