11.07.2015 Views

Untitled - UFRJ

Untitled - UFRJ

Untitled - UFRJ

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Bayesian Inference for Aggregated Functional Data withApplications to Electrical Data and ChemometricsNancy L. GarciaIMECC - Universidade Estadual de CampinasIn this work we address the problem of estimating mean curves when the available sample consists onaggregated functional data. Consider a typical curve for several sub-populations. Suppose that to samplefrom these individual curves is impossible (or too expensive). However, it is relatively easy to combinethese sub-populations and obtain sums of curves or weighted sums of curves. More specifically, replicatesof these curves are available and observations are made for “times” t=1,...,T. Our model specifies thatthe observed data is decomposed as the sum, over the C sub-populations, of latent structures whichare independent across sub-populations but temporally correlated. And these latent structures can bemodeled by Gaussian processes whose mean is a smooth curve depending on the sub-population andevolve with time t. Inference procedure is performed following the Bayesian paradigm. We apply ourmodel to a real dataset composed of the electric load of transformers which distributes energy to differenttypes of consumers and chemometric data obtained by Near-infrared (NIR) spectroscopy.Keywords: functional data, aggregated data, basis expansion, electric load monitoring, calibration,Near-infrared (NIR) spectroscopy14

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!