Generalization Error Bounds in Semi-supervised Classification Under the Cluster Assumption

摘要

We consider semi-supervised classification when part of the available datais unlabeled. These unlabeled data can be useful for the classificationproblem when we make an assumption relating the behavior of the regressionfunction to that of the marginal distribution. Seeger (2000) proposed thewell-known cluster assumption as a reasonable one. We propose amathematical formulation of this assumption and a method based ondensity level sets estimation that takes advantage of it to achieve fast ratesof convergence both in the number of unlabeled examples and the number oflabeled examples. color="gray">