Deep End-to-End One-Class Classifier

One-class classification (OCC) poses as an essential component in many machine learning and computer vision applications, including novelty, anomaly, outlier detection systems. With a known definition for target or normal set of data, one-class classifiers can determine if any given new sample spans within the distribution class. Solving this task general setting is particularly very challenging, due to high diversity samples from class absence supervising signal over novelty (nontarget) concept, which makes designing end-to-end models unattainable. In article, we propose adversarial training approach detect out-of-distribution trainable deep model. To end, jointly train two neural networks, R D. The latter plays discriminator while former, during training, helps D characterize probability by creating examples and, testing, collaborates with it novelties. Using our OCC, first test on image data sets, Modified National Institute Standards Technology (MNIST) Caltech-256. Then, several experiments video anomaly are performed University Minnesota (UMN) California, San Diego (UCSD) sets. Our proposed method successfully learn underlying outperforms other approaches.