Log-penalized linear regression
Julie Sweetkind-Singer · 2003
Regularization penalties are commonly used in linear regression to reduce overfitting (l). We introduce a log regularization penalty, motivated by a minimum-description-length (MDL) perspective (2) and from ideas in algorithmic complexity (3), and com- pare it to the more commonly used penalties known as ridge regreesion and the lasso (l). I. DISCUSSION