Estimating the Support of a High-Dimensional Distribution

The algorithm is a natural extension of the support vector algorithm to the case of unlabeled data by carrying out sequential optimization over pairs of input patterns and providing a theoretical analysis of the statistical performance of the algorithm.