How IBM Wants to Defend Neural Networks Against Adversarial Attacks

The security and robustness of deep neural networks(DNNs) architectures is one of the most important areas of research in the deep learning field. The native complexity of neural networks and its lack of interpretability makes them vulnerable to many forms of attacks. Some of the most sophisticated and scariest forms of attacks on DNNs are generated using other neural networks. Adversarial neural networks(ANNs) are often used to generate numerous attack vectors on DNNs by manipulating aspects such as the input dataset of the training policy. Protecting against adversarial attacks is far from being an easy endeavor as the attackers are always mutating and evolving.