Schedule:
15:00-15:30 coffee at the MCR
15:30-16:30 Seminar by Chris Holmes (Oxford)
Decision analysis and Bayesian inference using approximate models.
Bayesian statistics is founded in decision theory and many tasks of reasoning from data can be framed as aiding the selection of actions. For example, the optimal design of experiments or the optimal allocation of treatments to patients. Yet, formally, Bayesian statistics and statistical decision analysis are contingent on having the correct probability model for the data. Increasingly modern applications rely on statistical models that can scale at the expense of fidelity in order to deal with issues of data dimensionality and heterogeneity. A key question then arises as to the sensitivity of conclusions and the optimality of actions made when using approximate models? In this talk we present a framework for optimal decision-making under approximate models, illustrating the approach with motivating examples taken from the field of biomedical genetics.