查看文章

Enterprises are faced with various cyber security threats. In this dissertation, we focus on three aspects of cyber security threats: 1) cyber vulnerability prediction and patching, 2) cyber alerts management, and 3) review fraud detection. We propose a unified framework named Vulnerability Exploit Scoring and Timing (VEST) which predicts four aspects of cyber vulnerabilities, including 1) the probability being exploited, 2) the timing of vulnerability exploitation, 3) the over CVSS score and 4) the CVSS attributes of vulnerabilities. We then propose predictive patching, which uses a Markov Decision Process (MDP) model to reason about the uncertainty in the vulnerability predictions. We design a novel reinforcement learning based approach named BNB-DQN to suggest what to patch and when. To manage the overwhelming number of cyber alerts, we propose Predictive Cyber-Alert Management (PCAM). PCAM …