Proceedings of the annual belgian-dutch machine learning conference (Benelearn) 2008 pages:51-52
benelearn location:Spa, Belgium date:2008
In this paper we consider the problem of linear regression where some features might only be observable at a certain cost. We assume that this cost can be expressed in the same units and thus be compared to the approximation error cost. The learning task becomes a search for the features that contain enough information to warrant their cost.