The Science of Cybersecurity: Classification and Estimation with Human Actors and Adversaries

• Activity Number: 252
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Defense and National Security
• Author(s): Bronwyn Woods*+
• Companies: CMU/SEI/CERT
• Keywords: cybersecurity ; humans ; adversarial learning ; automation ; classification

Abstract

The security of cyber systems involves human actors in many roles. Defenders of a network must balance secure policies against productivity for users of the system. Creative adversaries innovate in response to any static defense strategy. The volume and dimensionality of network traffic demand automation, while the arms race between attacker and defender demands rapid adaptability and innovation. The Science of Cybersecurity (SoCS) team is researching solutions that augment human actors instead of replacing them with automation. This requires measurement and modeling of human performance, efficient elicitation of expert opinion and confidence, estimation of unobserved ground truth from multiple annotations, and principled integration of human actors with robust statistical prediction systems. I discuss these research areas with a focus on the opportunities for applying diverse statistical methodology in the rapidly evolving cybersecurity domain.

Copyright 2014 Carnegie Mellon University DM-0000928