This event is organised by the Programme on Computational Cosmology, an Oxford Martin School Programme
Speaker: Professor Mark Girolami, Professor of Statistics, Department of Statistical Science, UCL
Summary: The ability to perform Bayesian inference over a large class of problems has revolutionised the ability to reason under uncertainty in all the sciences from physics, engineering, biology, cosmology, economics, ecology to name a few. The ability to measure unprecedented volumes of data, ranging from time series, images, distributed acoustics, network interactions, protein concentrations places demands on our capability to efficiently analyse this data. Bayesian inference relies upon stochastic simulation methods such as Markov chain Monte Carlo which are being stretched to their limits in terms of the demands being placed on them by the sophisticated models and questions being asked in the sciences. This talk describes a way in which the current bottleneck in Markov chain Monte Carlo efficiency and capability can be alleviated by adopting a natural geometric view of the mechanistic models and associated probability distributions we wish to perform inference over. A number of illustrative examples in the biological and physical sciences will be presented allowing us to discuss the potential of this geometric perspective on simulation based Bayesian inference.
Biography: Mark Girolami is Professor of Statistics in the Department of Statistical Science. He also holds a professorial post in the Department of Computer Science at UCL and is Director of the Centre for Computational Statistics and Machine Learning (CSML), a large multi-faculty centre of world leading research excellence in statistical science, statistical learning theory, and statistical machine learning. His research, and that of his group, addresses the theory, methodology and application of Computational Statistics and has very strong multidisciplinary interactions with the life, clinical, physical and engineering sciences.