Skip to main navigationSkip to main content
The University of Southampton
The Care Life Cycle

(1) Activity and engagement effects on ageing cognition (2) Statistical methods for computer experiments Seminar

13:15 - 14:30
30 November 2012
Building 58, Room 4121

For more information regarding this seminar, please telephone Yvonne Richardson on 02380 598981 or email .

Event details

CLC Seminar

Abstract 1: Aston Research Centre for Healthy Ageing was set up to bring together Aston scientists across disciplines to work together on challenges in ageing research, based on the premise that we will only make further progress when we work together, applying a range of expertise in a complementary fashion. ARCHA's aims furthermore include working with policy makers, health service, community and business to have a real impact on older adults' lives, and specifically to include older adults in both research and impact planning. This talk will introduce ARCHA and a summary of how it all fits together. Given the centrality of cognition to successful ageing, for example, in terms of being able to plan and compensate for other functional losses, or in terms of predicting frailty and loss of independence, the talk will then review recent advances in prevention and rehabilitation of such losses. We will then summarise two ARCHA studies that aim to use these insights to work on predicting outcomes of physical, social and intellectual engagement interventions or behaviours on outcome quality of life and health indices.

Abstract 2: In this talk, we describe the application of statistical methods on the analysis of computer models. There is a large literature on computer experiments for deterministic models and we describe how it can be extended to stochastic systems. We motivate our methodology by tackling the problem of calibrating a stochastic traffic simulation model. Utilising fast surrogate models, known as emulators, we minimise the number of simulator runs required and speed-up the analysis.

We describe heteroscedastic emulators where the simulator response is assumed to be normally distributed but the variance is allowed to depend on the inputs. We present a model-based experimental strategy to optimally learn the parameters of a heteroscedastic emulator. For cases where the normality assumption is too restrictive, we present the quantile emulator where quantiles of the response are directly modelled.

We conclude by demonstrating the utilisation of the emulators in a calibration analysis of a stochastic traffic simulator model. Using an iterative design approach, we show how the parameters that lead to plausible simulator output are progressively better identified.

Speaker information

Dr Carol Holland, Aston Research Centre for Healthy Ageing (ARCHA). Director of ARCHA

Dr Alexis Boukouvalas, Aston Research Centre for Healthy Ageing (ARCHA). Research Fellow at ARCHA

Privacy Settings