Reconciling malware labeling discrepancy via consensus learning

摘要: Anti-virus systems developed by different vendors often demonstrate strong discrepancy in the labels they assign to given malware, which significantly hinders threat intelligence sharing. The key challenge of addressing this stems from difficulty re-standardizing already-in-use systems. In paper we explore a non-intrusive alternative. We propose leverage correlation between malware anti-virus create “consensus” classification system, through can share information without modifying their own labeling conventions. To end, present novel integration framework Latin exploits correspondence participating as reflected heterogeneous at instance-instance, instance-class, and class-class levels. provide results extensive experimental studies using real datasets concrete use cases verify efficacy reconciling discrepancy.