Defending Hardware-Based Malware Detectors Against Adversarial Attacks

In the era of Internet Things (IoT), Malware has been proliferating exponentially over past decade. Traditional anti-virus software are ineffective against modern complex Malware. order to address this challenge, researchers have proposed hardware-assisted detection (HMD) using hardware performance counters (HPCs). The HPCs used train a set machine learning (ML) classifiers, which in turn, distinguish benign programs from Recently, adversarial attacks designed by introducing perturbations HPC traces an sample predictor misclassify program for specific HPCs. These with basic assumption that attacker is aware being detect Since processors consist hundreds HPCs, restricting only few them aids attacker. article, we propose moving target defense (MTD) attack designing multiple ML classifiers trained on different sets MTD randomly selects classifier; thus, confusing about or number applied. We developed analytical model proves probability guess perfect HPC-classifier combination extremely low (in range $10^{-1864}$ system 20 HPCs). Our experimental results prove able improve classification accuracy modified through generator up 31.5%, near (99.4%) restoration original accuracy.