Title: Bayesian modelling of the selection bias problem in regression
Authors: M Remedios Sillero-Denamiel - Trinity College Dublin (Ireland) [presenting]
Simon Wilson - Trinity College Dublin (Ireland)
Hieu Cao - School of Computer Science and Statistics-Trinity College Dublin (Ireland)
Abstract: In the regression setting, it is typically assumed that training and test sets follow similar distributions, but that is not always true, as is the case with the sky surveys of galaxies where faint ones are not observed in favour of brighter ones. In addition, when data follow complicated non-Gaussian distributions, the full conditional density has to be estimated to properly quantify the uncertainty in the predictions. We present a Bayesian approach to estimate the conditional density under selection bias.