Reinforcement Learning for Zero-Day Vulnerability Detection in IoT Devices: A Proactive Approach

Abstract

Zero - Day vulnerabilities pose a significant threat to the security of IoT devices, as they remain undetected and unpatched by vendors. In this research paper, we propose a novel approach to Zero-Day Vulnerability Detection in IoT devices using reinforcement learning for conjecture generation. Our model leverages real-time telemetry data from IoT devices and metadata about the network to generate potential conjectures about Zero-Day vulnerabilities. The agent is trained with a deep reinforcement learning architecture and benefits from a human-in-the-loop to incorporate domain expertise. We evaluate the performance of the model on a real-world IoT testbed and compare it with existing approaches, demonstrating its proactive and efficient nature in detecting Zero-Day vulnerabilities. Keywords: Zero-Day Vulnerabilities, IoT Devices, Reinforcement Learning, Conjecture Generation Vulnerability Detection