One World ABC Seminar: Marko Järvenpää

on the October 1, 2020

at 11:30 am [UK time]
For this eleventh session of the One World ABC Seminar, Marko Järvenpää from the university of Oslo will talk about "Batch simulations and uncertainty quantification in Gaussian process surrogate ABC".

Inspired by the "One World Probability Seminar", we decided to run The One World ABC Seminar, a weekly/fortnightly series of seminars that will take place on Blackboard Collaborate on Thursdays at 11.30am [UK time]. The idea is to gather members and disseminate results and innovation during these weeks and months under lockdown.
 


Marko Järvenpää

Abstract
The  computational  efficiency  of  approximate  Bayesian  computation  (ABC)  has  been  improved by using surrogate models such as neural networks and Gaussian processes (GP). In one such promising framework the discrepancy between the simulated and observed data is modelled with a GP surrogate which is further used to form a model-based estimator for the intractable posterior and to select new simulation  locations adaptively so as to maximise sample-efficiency. In this talk we show how to improve this approach in several ways. Most importantly, we develop batch-sequential Bayesian experimental design strategies to parallelise the expensive simulations. In earlier related work only sequential strategies have been used. Current surrogate-based ABC methods also do not fully account the uncertainty due to the limited budget of model simulations as they output only a point estimate of the ABC posterior. We propose a numerical method to fully quantify the uncertainty in, for example, ABC posterior moments. We call the resulting improved framework as “Bayesian ABC”and provide a detailed discussion on its connection to Bayesian optimisation, Bayesian quadrature and GP-based level set estimation methods. Experiments with toy and real-world simulation models are used demonstrate advantages of the proposed techniques.
This is joint work with Aki Vehtari and Pekka Marttinen. This work was done while the speaker was with Department of Computer Science, Aalto University, Finland.
 
References
M Järvenpää, A. Vehtari and P. Marttinen. Batch simulations and uncertainty quantifi-cation in Gaussian process surrogate approximate Bayesian computation. Proceedings ofthe 36th Conference on Uncertainty in Artificial Intelligence (UAI), PMLR volume 124,http://proceedings.mlr.press/v124/jarvenpaa20a.html
 
Published on September 28, 2020