Perceiving Behavior of Cyber Malware with Human-Machine Teaming

Cyber malware has evolved from simple hacking programs to highly sophisticated software engineering products. Human experts are in high demand but are busy, expensive, and have difficulty searching through massive amount of data to detect malware. In this paper, we develop algorithms for machines to learn visual pattern recognition processes from human experts and then to map, measure, attribute, and disrupt malware distribution networks. Our approach is to combine visualization and machine vision for an intuitive discovery system that includes visual ontology of textures, topological structures, traces, and dynamics. The machine vision and learning algorithms are designed to analyze texture patterns and search for similar topological dynamics. Compared to recent human-machine teaming systems that use input from human experts for supervised machine-learning, our approach uses fewer samples, i.e. less training, and aims for novel discoveries through human-machine teaming.