12 August 2020 | 08:00 UTC/GMT
Time Zone Converter
In this big data era, the major trends of currently available data sets are larger, more heterogeneous and more complex. Indeed, temporal data are routinely collected in various sectors (e.g. biology, environment, health, socioeconomics). Some health data are scalar (e.g. genetic); lifestyle data (e.g. diet, physical activity) are longitudinal; pollution and weather data are functional. However, to extract the wealth of information included in these data sets is challenging due to the lack of statistical methodology for dimension reduction and modeling which addresses these trends.
Functional Data Analysis (FDA), a branch of statistics that anayses data providing information about curves, surfaces or anything else varying over a continuum, has established itself as an important and dynamic field. It offers effective new tools and has developed new methods and theories to analyse both sparse and dense temporal and spatial datasets. For instance, Functional Principal Component Analysis (FPCA), Functional Regression Analysis (FRA) etc. With the advent of more and more complex and dynamic data in Biometrics and Health Sciences, the use of FDA methods will increase. For example modeling the effects of pollutants and lifestyles on hospital admissions, modeling the effect of temperature and humidity changes on tree growth and using information from wearable devices which produce data on an individuals’ health constantly. State of the art methods for modeling functional data are, for instance, FPCA the main aim of which is to carry out dimension reduction, FRA which contains a huge number of approaches for investigating the relationship (linear) between response and predictors. However, currently available datasets requires new developments. Examples include how to deal with the heterogeneous predictors (scalar or functional), how to test the efficiency of dimension reduction simultaneously (response and predictors), how to characterize the correlation among these heterogeneous predictors, how to classify the subjects based on the prediction results, etc. Thus there is a high demand for efficient and flexible statistical approaches to model functional data.
In this session, we will bring together internationally recognized statisticians (Lijian Yang), biostatisticians (Haiyan Liu, Jeanine Houwing-Duistermaat) and machine learners (Theo Damoulas) with expertise in modeling dense and sparse temporal data. The confirmed speakers will share their novel findings on how to estimate the effect lag of predictors in functional linear model and do prediction with application to survival of trees in Amazonian rainforest (Haiyan), how to model non-stationary multi-scale spatial-temporal processes with application in air quality in London (Theo), the estimation and inference for spatial-temporal data (Lijian).
Jeanine Houwing-Duistermaat, University of Leeds
Haiyan Liu, University of Leeds
On estimation of the effect lag of predictors and prediction in functional linear model
Lijian Yang, Tsinghua University
Prediction of working memory ability based on EEG by functional data analysis