FURI | Spring 2021

Developing Robust Defenses for Deep Neural Networks

As the use of deep learning in computer vision becomes integrated in many technologies including self-driving cars, electronic banking, and security systems, attackers are seeking to exploit weaknesses that are imperceptible to the human eye, but trick computers. This research explores more robust methods to train deep neural networks so that they are resistant against such attacks in order to keep people safe from harm. Further research needs to be done to find new attack techniques in order to verify the effectiveness of the defenses.