The talks of the Analysis workshop will be held in room S1.06 of the Strand Building.
Tea & Coffee will be offered in the Bush House Arcade.
Invited speakers
Tuesday 7th
Advances in Computational Statistics 14.15  15.00: Anthony Lee
15.00  15.45: Matias Quiroz 15.45  16.30: Tea & Coffee [BH(N) Arcade] 16.30  17.15: Nicolas Chopin 17.15  18.00: Arnaud Doucet 
Wednesday 8th
Design and analysis of experiments 14.15  15.00: Stefanie Biedermann
15.00  15.45: Hugo MaruriAguilar 15.45  16.40: Tea & Coffee [BH(N) Arcade] 16.40  17.20: Ben Parker 17.20  18.00: Daria Semochkina 
Anthony Lee
Solving the Poisson equation using coupled Markov chains Abstract. For a timehomogeneous Markov chain, the solution of the Poisson equation, g, associated with some function h can be used to analyze ergodic averages of h. Despite its importance as a theoretical device, it is not easy to approximate it for practical Markov chains that are used for statistical estimation. We demonstrate that for a wide variety of Markov chains, the solution of the Poisson equation can be approximated pointwise in an unbiased manner using coupled Markov chains. This leads naturally to unbiased and/or consistent approximations of the asymptotic variance in the Markov chain CLT. 
Stefanie Biedermann
Designing experiments when data may be missing not at random Abstract. Missing data are a wellknown problem in the process of data collection. The effects are wide ranging, and the loss of data can lead to inefficiencies and introduce bias into analyses, in particular when data are missing not at random (MNAR). Recovering missing values through a follow up sample can mitigate the effect of MNAR missing data, and in particular allows researchers to conduct hypothesis tests for MNAR missingness, which would not be possible when using only the original incomplete data. In this talk, we explore strategies for designing follow up samples to test for MNAR in an efficient and costeffective way to ensure the practicality of this approach.

Matias Quiroz
Spectral approaches to speed up Bayesian inference for large stationary time series data Abstract. This talk will discuss some recent approaches to speed up MCMC for large stationary time series data via data subsampling. We discuss the Whittle loglikelihood for univariate time series and some properties that allow estimating the loglikelihood via data subsampling. We construct an efficient estimator using grouped quadratic control variates that are more robust to the assumption of an approximately quadratic logdensity that underpins the approach using standard nongrouped quadratic control variates. Finally, we consider an extension to multivariate time series via the multivariate Whittle loglikelihood. 
Hugo MaruriAguila
Hilbert series, Smolyak grids and a matrix identity Abstract. This talk is about continuing work on the analysis of Smolyak grids. These grids are experimental designs with a reduced number of runs (“sparse grids”) that still preserve the accuracy of approximations. In particular, the asymptotic accuracy of the method remains equal while using fewer points. Our work leads to efficient construction of interpolators for Smolyak grids, and uses some nice results from algebra to formulate an inclusionexclusion formula for interpolation. As part of our developments, we show that the inverse of the hierarchical design model matrix satisfies an identity that is precisely the same inclusionexclusion principle. 
Nicolas Chopin
Wastefree SMC Abstract. (Joint work with HaiDang Dau) A standard way to move particles in a SMC sampler is to apply several steps of a MCMC (Markov chain Monte Carlo) kernel. Unfortunately, it is not clear how many steps need to be performed for optimal performance. In addition, the output of the intermediate steps are discarded and thus wasted somehow. We propose a new, wastefree SMC algorithm which uses the outputs of all these intermediate MCMC steps as particles. We establish that its output is consistent and asymptotically normal. We use the expression of the asymptotic variance to develop various insights... on how to implement the algorithm in practice. We develop in particular a method to estimate, from a single run of the algorithm, the asymptotic variance of any particle estimate. We show empirically, through a range of numerical examples, that wastefree SMC tends to outperform standard SMC samplers, and especially so in situations where the mixing of the considered MCMC kernels decreases across iterations (as in tempering or rare event problems). 
Ben Parker
Design of experiments for networks, and networks for experimental design
Abstract. How can we design experiments on networks, and how can ideas from network science help us to find designs?

Arnaud Doucet
Diffusion Schrodinger Bridges: From Generative Modeling to Inference Abstract. Denoising diffusion models, also known as scorebased generative models, have recently emerged as a powerful class of generative models. They provide stateoftheart results, not only for unconditional simulation, but also when used to sample from complex posterior distributions arising in a wide range of inverse problems such as image inpainting or deblurring. A limitation of these models is that they are computationally intensive as obtaining each sample requires simulating a nonhomogeneous diffusion process over a long time horizon. ... We show here how a a Schrodinger bridge formulation of generative modeling leads to a theoretically grounded algorithm shortening generation time which is complementary to other proposed acceleration techniques. We further extend the Schrodinger bridge framework to perform posterior simulation. We demonstrate this novel methodology on various applications including image superresolution and optimal filtering for statespace models. 
Dasha Semochkina
Efficient emulation of epidemiological spatiotemporal patch models Abstract. Modern computational science allows complex scientific processes to be described by mathematical models implemented in computer codes, or simulators. When these simulators are computationally expensive, it is common to approximate them using statistical emulators constructed from computer experiments. Often, the simulator output represents system behaviour across a large spatial and/or temporal domain, which can make efficient emulation computationally challenging. Epidemic progression is an important application area... utilising temporal (and often spatial) simulation. Our motivating application simulator is a multipatch compartmental model of flu, with each patch representing a region of Botswana. We have extended the methodologies of Outer Product Emulators (OPEs) and Parallel Partial Emulators (PPEs) to emulate the epidemiological model for the spread of flu through a spatially dispersed population. Specifically, the model produces time series of the number of infected people in different patches of the space, making the output spatiotemporal. We compare and analyse the benefits and drawbacks of each emulation approach in application to the flu model. 