Reinforcement-GAN for Network Intrusion Detection

Network intrusion detection systems play a crucial role in computer networks defense. However, most of the existing solutions do not take into account the time-evolving component of the attacks. Attackers are constantly changing the intrusion methods which becomes a serious threat and requires the development of adaptive methods that can cope with these modifications. Moreover, the lack of labeled data or the delay in obtaining labels is also an important challenge.
We propose a system, Reinforcement-GAN, that is able to adapt to new network intrusion methods by incorporating reinforcement learning and generative adversarial networks (GANs). Our system takes advantage of reinforcement learning techniques for allowing the development of a more adaptive model building while utilizing GANs to improve its detection accuracy. Our system follows a semi-supervised method that integrates a self-learning approach while updating the model training with labeled data as it becomes available.

Faculty Supervisor:

Paula Branco

Student:

Partner:

Tecnologico de Monterrey, Campus Toluca

Discipline:

Computer science

Sector:

Artificial Intelligence; Information and Communications Technology

University:

University of Ottawa

Program: