Using Deep Reinforcement Learning to Simulate Security Analyst

The goal of this project was to overcome conventional limitations of deep learning approaches and advance the use of machine learning approaches to understand, identify, and analyze security incidents. In general, main-stream classifiers require a full description of the sample (in our scenario this means all possible information about the security incident) and perform the classification in one step, which is in a sharp contrast to the modus operandi of the analyst, whose investigation is composed of a sequence of actions and decisions.