A Bayesian Model Combination-based approach to Active Malware Analysis

Abstract

Active Malware Analysis involves modeling malware behavior by executing actions to trigger responses and explore multiple execution paths. One of the aims is making the action selection more efficient. This paper treats Active Malware Analysis as a Bayes-Active Markov Decision Process and uses a Bayesian Model Combination approach to train an analyzer agent. We show an improvement in performance against other Bayesian and stochastic approaches to Active Malware Analysis.