R2 and adjusted R2 may exaggerate a model's true ability to predict the dependent variable in the presence of overfitting, whereas leave-one-out R2 (LOOR2) is robust to overfitting. We demonstrate this by replicating 279 regressions from 100 papers in top economics journals, where the median...

This article describes how to use heterogeneous information in speech enhancement. In most of the current speech enhancement systems, clean speeches are recovered only from the signals collected by acoustic microphones, which will be greatly affected by the acoustic noises. However,...