Learning Under Non-stationarity: Covariate Shift Adaptation by Importance Weighting

摘要: The goal of supervised learning is to estimate an underlying input-output function from its training samples so that output values for unseen test input points can be predicted. A common assumption in the follow same probability distribution as points. However, this not satisfied, example, when outside region extrapolated. situation where and different distributions while conditional given unchanged called covariate shift. Since almost all existing methods assume are drawn distribution, their fundamental theoretical properties such consistency or efficiency no longer hold under In chapter, we review recently proposed techniques shift adaptation.