Towards Secure and Reliable Deep Learning Systems against Adversarial Attacks

In recent years, deep learning methods have been widely deployed in a range of vision-related tasks such as object detection, segmentation and recognition. However, such methods can be vulnerable to adversarial attacks that subtle perturbations to inputs can result in incorrect decisions. In this research, we attempt to explore the new generation of adversarial attacks, improve the adversarial robustness of deep neural networks and establish reliable deep learning systems against adversarial attacks for secure digitalization and smart society. This research is also expected to have a great practical and social impact due to the wide applicability of automatic systems to our daily life. This research includes both theoretical analysis and experimental validations using publicly available datasets. Mainstream computer vision and machine learning methods will also be investigated. The research will be carried out in the Center for Machine Vision and Signal Analysis, University of Oulu.

Strategic research project of the University of Oulu

Focus institute: Infotech Oulu

Faculty: Faculty of Information Technology and Electrical Engineering (ITEE)