Department of Computer Engineering
ROBUST DEEP ONE-CLASS KERNEL SPECTRAL REGRESSION
(Supervisor: Asst. Prof. Dr. Shervin R. Arashloo)
Computer Engineering Department
One class classification also known as OCC is a special case of multi-class classification where data observed during training is from a single positive class. The main aim of OCC is to learn a representation and/or classifier that enables recognition of positive labeled queries during inference. In recent years, this field has gained considerable amount of interest especially in areas in Computer vision, Machine Learning and Biometrics where data from other classes are difficult/impossible to obtain. It is being used in Anomaly detection, Novelty detection, Outlier detection as well as medical imaging. Typically people have been using traditional Machine learning algorithms like One-class SVM. However, with the advent of deep learning, this area has improved considerably. In our work, we show that an end-to-end designed deep Convolutional Neural Network with the benefits of Kernel Hilbert space is able to beat the state-of-the-art algorithms in this domain. Moreover, we also hope to show that our algorithm is robust to contamination in the training data and is still able to outperform baseline kernel null-space methods as well as other existing approaches in the OCC paradigm.
DATE: 12 April 2021, Monday @ 15:30