Multitask Learning Using Partial Least Squares Method

In the machine learning field, feature selection is used to discard the redundant information and improve the learning accuracy. In this paper, the redundant information is reused in the learning of partial least squares method within the frame of multitask learning. This newly proposed method is used to solve the multivariate calibration problem, a classic problem in the analytical chemistry field. Results on three data sets collected using fluorescence spectroscopy show that multitask learning can help to improve the prediction accuracy of partial least squares method greatly.