MTLEARN Procedure

The MTLEARN procedure implements the multitask learning technique for least squares loss with and graph structure penalizations. It solves multiple related sparse linear regression problems simultaneously. A graph structure encodes the relationships between the problems. PROC MTLEARN shares the data and model parameters among different regression problems and solves the problems simultaneously in order to produce a more robust and accurate predictive model. When used with independent graph tables, PROC MTLEARN also serves as an elastic net regularization solver, which is widely used in feature selection.