Confidence Intervals for Maximin Effects in Inhomogeneous Large-Scale Data

摘要

One challenge of large-scale data analysis is that the assumption of anidentical distribution for all samples is often not realistic. An optimallinear regression might, for example, be markedly different for distinct groupsof the data. Maximin effects have been proposed as a computationally attractiveway to estimate effects that are common across all data without fitting amixture distribution explicitly. So far just point estimators of the commonmaximin effects have been proposed in Meinshausen and B\"uhlmann (2014). Herewe propose asymptotically valid confidence regions for these effects.