Skip to main content

Table 2 Comparison of the existing adversarial defense approaches in malware detection

From: LSGAN-AT: enhancing malware detector robustness against adversarial examples

Ref

Defense mechanism

Data type

Data level

Model

Robustnes analysis

Components analysis

AME generation

AME evaluation

Li and Li 2020)

Proactive

Drebin, Androzoo

SC, Byte

Ensemble Methods (DL-based)

Grosse et al. 2016)

Proactive

Drebin

SC

FFNN

Chen et al. 2017)

Reactive

Windows

SC

SecDefender (ML-based)

Grosse et al. 2017)

Proactive

Drebin

SC

DNN

Khoda et al. 2019)

Proactive

Drebin

SC

Ensemble Methods (ML-based)

Sewak et al. 2020)

Proactive

MALICIA (Nappa et al. 2015)

Opcode

Deep reinforcement learning

Wang, et al. 2017)

Reactive

Window Audit Log (Berlin et al. 2015), MINST, CIFAR-10

SC, Pixel (image)

DNN

Ours

Proactive

VirusShare, Androzoo

SC

LSGAN

  1. aThe Defense Mechanism includes proactive and reactive. The Robustness Analysis aims to evaluate whether the defense model can recognize the AME of other attack models. The Component Analysis is used to evaluate whether the reference paper makes components analysis. The AME generation exposes whether the defense model can generate AME, and normally, a proactive defense mechanism will generate the AME