PHD Gans for outlier Detection

The project aims at researching a novel anomaly detection model, namely it will address the problem of determining normal from abnormal observations when the dataset in hand is highly biased towards the normal class. This will be achieved by using modern deep learning frameworks (generative adversarial networks) in combination with traditional anomaly detection algorithms such as one-class support vector machines. Applications will include the reference benchmark MNIST and CIFAR image datasets, along with more complex brain imaging data related to different mental health conditions.