A0833
Title: Importance weighted orthogonal greedy algorithm with estimated weight function
Authors: Shinpei Imori - Hiroshima University (Japan) [presenting]
Ching-Kang Ing - National Tsing Hua University (Taiwan)
Abstract: Greedy-type algorithms are feasible for prediction in high-dimensional linear regression models when the number of explanatory variables is larger than the sample size. A greedy algorithm is studied under the covariate shift, where the distribution of explanatory variables in training data possibly differs from that in test data. The proposed algorithm needs to use an unknown weight function based on the density ratio. A sufficient condition is given in order that the proposed algorithm with the estimated weight function archives a good convergence rate.
