Presentation
Stochastic-HMDs: Adversarial-Resilient Hardware Malware Detectors via Undervolting
DescriptionMachine learning-based hardware malware detectors
(HMDs) offer a potential game changing advantage in defending
systems against malware. However, HMDs suffer from adversarial
attacks, can be effectively reverse-engineered and subsequently
be evaded, allowing malware to hide from detection. We address
this issue by proposing novel HMDs (Stochastic-HMDs), which
leverage approximate computing to harden HMDs against
adversarial evasion attacks. Stochastic-HMDs introduce stochastic
noise into the computations within the model, through undervolting, to build an efficient
and low-cost moving-target defense. We demonstrate that Stochastic-HMDs offer effective
defense against adversarial attack, by detecting 94% evasive malware with <2% accuracy loss and 15% power savings.
(HMDs) offer a potential game changing advantage in defending
systems against malware. However, HMDs suffer from adversarial
attacks, can be effectively reverse-engineered and subsequently
be evaded, allowing malware to hide from detection. We address
this issue by proposing novel HMDs (Stochastic-HMDs), which
leverage approximate computing to harden HMDs against
adversarial evasion attacks. Stochastic-HMDs introduce stochastic
noise into the computations within the model, through undervolting, to build an efficient
and low-cost moving-target defense. We demonstrate that Stochastic-HMDs offer effective
defense against adversarial attack, by detecting 94% evasive malware with <2% accuracy loss and 15% power savings.
Event Type
Research Manuscript
TimeWednesday, July 12th11:25am - 11:40am PDT
Location3003, 3rd Floor
AI
RISC-V
AI/ML Application and Infrastructure