When: Thursday 14th September 2017, 13:00-14:00.
Included in meeting:
This week we have special presenters Richard Boys and Stephen Johnson of Newcastle University presenting.
They will be in the office for the remainder of the week and fly back to the UK on the 18th September.
Details on their presentations are included below.
“The Plackett-Luce model and its “choice order problem”
Abstract: Ranked data arises in many areas of application, such as sport, politics and marketing. Formally, a ranking can be viewed as a permutation of a set of entities. Many of the models for ranked data rely on strong assumptions about how each ranker forms their ranking, that is, the order in which they assign positions in their ranking. This talk considers an extension of the popular Plackett-Luce model which relaxes the known choice order assumption and instead infers the ordering from the rankings themselves. There are many challenges in determining the posterior distribution, particularly in exploring permutation space. However, as we shall see, the problem of poor mixing MCMC chains can be overcome through the use of suitable proposal distributions and parallel tempering. ”
“Efficient construction of optimal designs for stochastic process models
Abstract: Stochastic process models can be very useful in monitoring (dynamic) experimental scenarios. Often changes to the underlying system are detected by changes to the model parameters and so efficient sampling schemes are needed to detect such changes. We look at a new method to determine Bayes optimal (sampling) designs in a fast and computationally efficient way.”
Next BRAG (28/09) we will have Miles and Mahasen presenting.
We look forward to seeing you there!
Matt and Shovanur