Important deadlines:

Submission (closed)
May 11, 2014
Acceptance notification
June 15, 2014
Payment of fee
July 10, 2014

Last modified:
December 27, 2021 15:49:06 (gk)

Chris Holmes: Foundations of Bayesian inference with special emphasis on the use of approximate models (Keynote Talk C, Friday, 9:00)

Bayesian inference is founded in decision theory following the work of Savage. However, formally, Bayesian analysis assumes that the form of the true model (sampling distribution) is known and has support under the prior, and that the prior is precisely calibrated to reflect subjective beliefs. In practice both the prior and the sampling distribution (likelihood) are approximations. Moreover, in an era of "big-data" and highly heterogeneous unstructured data there is an increasing necessity for computationally tractable approximate models, which are misspecified by design. But what is the justification for then updating via Bayes theorem, and what are the consequences and sensitivity of conclusions to model misspecification? In this talk we will review the decision theoretic foundations of Bayesian inference and discuss recent work to deal formally with the issue of approximations and the sensitivity of inference to model misspecification.