This work will focus on three areas of theoretical statistics: (1) Bayesian modelling and analysis of time series, including time domain and frequency domain approaches to both long memory and short memory models, (2) other Bayesian projects including nonlinear regression models and measurement error models, and (3) Bayes causal networks and belief functions, including mathematical characterization of these new models, as well as the development of principles for model construction and computation that are required for practical use. Dempster's work in theoretical statistics is becoming foundational for other scientists attempting to incorporate the realities of uncertainty into expert computer systems. In addition to extending relevant theory, he will tackle the computational bottleneck that now blocks successful widespread implementation of these models.
