Mixture of Support Vector Data Descriptions
Vinh Lai, Duy Nguyen, Khanh Duy Tung Nguyen, Trung Bao. Le · 2015
We present Mixture of Support Vector Data Descriptions (mSVDD) for one-class classification or novelty detection. A mixture of optimal hyperspheres is automatically discovered to describe data. The model consists of two parts: log likelihood to control the fit of data to model (empirical risk) and regularization quantizer to control the generalization ability of model (general risk). Expectation Maximization (EM) principle is employed to train the model. We demonstrate the advantage of the proposed model: if learning mSVDD in the input space, it simulates learning single hypersphere in the feature space and the accuracy is thus comparable but the training time is significantly shorter.