Kaushik Sinha and Mikhail Belkin (Spotlight ID M22) _____________________________________________________  Classification with labeled and unlabeled examples
­

The Value of Labeled and Unlabeled Examples when the Model is Imperfect

Classical Result for Parametric Model




Model parameters can be estimated polynomially fast in the number of unlabeled examples  Using small number of labeled examples, class attributes can be identified exponentially fast in the number of labeled examples Question
­

What roles do the labeled and unlabeled examples play when the true model is only slightly different from the parametric model?