One World ABC Seminar: David Frazier

on the October 15, 2020

at 10:30 am [UK time]
For this twelfth session of the One World ABC Seminar, David Frazier from Monash University will talk about "Robust and Efficient Approximate Bayesian Computation: A Minimum Distance Approach".

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.

David Frazier

Approximate Bayesian computation (ABC) is now routinely applied to conduct inference incomplex models.  However, there remain at least two significant hurdles to the widespreadadoption of ABC methods in practice.  First, since ABC replaces the observed sample withsummary  statistics  and  the  likelihood  function  with  a  given  metric  (for  the  summaries),ABC-based inference inevitably entails a loss in statistical efficiency.  Second, the choice ofsummaries and metric in ABC ensures that, as a general method, ABC may not be robustto deviations from the underlying model structure:  different summaries/metrics can lead tosignificantly different inferences under model misspecification.  Motivated by these efficiencyand robustness concerns, we construct a new approximate Bayesian inference approach thatdelivers point estimators that are as efficient as those obtained by exact Bayesian inference,even when the latter is infeasible to implement, while also simultaneously displaying robust-ness to deviations from the underlying model assumptions.  Several examples demonstratethat this new approach outperforms state of the art approaches to ABC-based inference, andcompares favorably with exact Bayes inference in correctly-specified models, while outper-forming these approaches when the model is misspecified.
[1]  D.T. Frazier (2020). Robust and Efficient Approximate Bayesian Computation:  A Min-imum Distance Approach, arXiv:2006.14126
Published on October 7, 2020