Multi-level network Lasso for multi-task personalized learning

Jiankun Wang, Luhuan Fei, Lu Sun

Abstract: Highlights•Multi-level network Lasso is proposed to group samples based on feature subsets.•It is extended to multi-task scenarios by learning task groups in feature subspace.•We generalize it based on lp quasi-norm to avoid over-penalization on outliers.•Its effectiveness is demonstrated on synthetic and real-world datasets.