Generalization bounds for learning with linear, polygonal, quadratic and conic side knowledge (Q747246)

The authors consider the problem of supervised learning with side knowledge on labels of unlabeled examples. The side knowledge has the effect of reducing the hypothesis space, leading to tighter generalization bounds, and thus possibly better generalization. In this paper the authors study three types of side knowledge, one with linear and polygonal constraints on the hypothesis space, another with quadratic constraints, and the last with conic constraints. Bounds on complexity measures of the hypothesis space are provided.