A Sp ectral Regularization Framework for Multi-Task Structure Learning
A. Argyriou, C. A. Micchelli, M. Pontil, Y. Ying ID W7

minimize

Err(W ) +  Penalty(W, D)

Alternating minimization algorithm · Learn tasks' parameters W (e.g. using SVMs) · Learn structure matrix D as a spectral function of W (often has closed-form solution)