RSS Merseyside Local Group: Bayesian Real-Time Modelling скачать в хорошем качестве

RSS Merseyside Local Group: Bayesian Real-Time Modelling

As the nature of the world around us grows increasingly digitised in the form of large-scale data, new pathways open up to understanding and responding to global issues. Combined with Bayesian approximation, dynamic input data make for a powerful approach to real-time analysis that can, for example, predict flooding risks or evaluate disease control. These talks will explore theory and applied cases of Bayesian modelling in real-time. ---------------------------------------------------------------------------------------------------------------- Prof. Darren Wilkinson (Durham University) – “Streaming data modelling for real-time monitoring and forecasting” A key challenge of modern data science is the development of scalable algorithms for extracting useful information from large, complex, heterogeneous and ever-growing data sets in (near) real-time. A promising approach is to combine sequential statistical modelling with computational frameworks for streaming data processing, making significant use of functional programming approaches. This talk will introduce some of the basic concepts, and illustrate the techniques in the context of real-time urban flood forecasting. Fusion of highly non-Gaussian point and areal spatial data from environmental sensors provides a challenging use case for on-line statistical modelling. This work was supported by UKRI-NERC and The Alan Turing Institute. ---------------------------------------------------------------------------------------------------------------- Prof. Peter Neal (University of Nottingham) – “Statistical inference for emerging diseases inspired by Covid-19” The Covid-19 pandemic has transformed the landscape in terms of how mathematical modelling of infectious diseases is viewed both in the academic community and wider population and has provided a stimulus for much research. I will talk about one of the research directions, motivated by Covid-19, which I have been exploring. We start from the question: What does the (partial) observation of the times at which individuals show symptoms to a disease tell us about the number of individuals infected in a population? For a general stochastic epidemic (SIR) model we are able to answer this question. By using a birth-death process approximation we can obtain explicit distributions for the number of people infectious when the infection and recovery rates and probability of detection of cases all vary over time. The key consequence of this, from a statistical perspective, is that it yields a likelihood for the observation process of detection times of cases without having to impute infection times. This enables us to: Provide fresh insight into the likelihood of ongoing epidemics with the results supporting a Bayesian approach to analysing such epidemic models. Through a fast approximation of the likelihood analyse European Covid-19 data and evaluate the effectiveness of control measures.