S3RI Seminar - "Efficient model comparison techniques for models requiring large scale data augmentation", Professor Peter Neal (Lancaster University) Seminar
For more information regarding this seminar, please email Dr Helen Ogden at H.E.Ogden@southampton.ac.uk .
Event details
Selecting between competing statistical models is a challenging problem especially when the competing models are non-nested. We offer a simple solution by devising an algorithm which combines MCMC and importance sampling to obtain computationally efficient estimates of the marginal likelihood (model evidence) which can then be used to compare the models. The incorporation of a particle filter step into the algorithm allows the algorithm to be successfully applied to time series data sets, where calculating the marginal likelihood is made more challenging by the presence of large amounts of missing data.
Speaker information
Professor Peter Neal, Lancaster University. PhD Supervision Interests cover: The understanding and control of infectious diseases is of considerable importance to society. How a disease spreads and/or how infectious a disease is, has tremendous implications upon the health and wealth of a community. I am interested in both the probabilistic and statistical analysis of infectious diseases. From a probabilistic perspective, we look to answer questions as: What is the probability that a disease takes hold within a community? How many individuals are ultimately infected by the disease? This involves developing novel probabilistic techniques to answer these questions for realistic population models such as the household and random graph models. Alternatively, having observed an epidemic we can propose a model for the disease spread and estimate the model parameters. However, often the disease data are "incomplete" and novel statistical methods, in particular, Markov Chain Monte Carlo (MCMC) are required to analyse the data. We aim to answer questions concerning the adequacy of the model and the predictive capabilities of the model for the future epidemic outbreaks.