# Adversarial Attack Approaches to Measure the Robustness of IoT Malware Detectors * 論文的論述: Adversarial attack vector 惡意程式每天上百萬支變異產生,越來越多ML-BASE的檢測器,且偵測率越來越高。 針對ML-BASE的檢測器進行分析,有效量測量化IoT Detector強韌性。(反思ML-BASE的檢測器的不足) * 對抗式攻擊的方法: 子揚攻擊透過4種generate four types of attack opcodes while preserving the original functionality(attack vector),Binary的方式Inject payload泛用性更高。 * 攻擊演算法: Hill climbing algorithm With 4 types: Single block、Loop、Transform opcode、Arithmetic opcode 需要額外增加Opcode 打法論述,對於後面的Detector的評估面向比較完整 * 實驗的比較 detector accuracy / attack success rate / number of attacks achieve 疊代次數多少會達0.5: | Detector | ML | | | DL | | | | | -------- | ------- | ------------- | --- | --- | ---- | --- | --- | | | XGBoost | Random Forest | SVM | DNN | LSTM | CNN | RNN | | Opcode | | | | | | | | | n-gram | | | | | | | | | Opcode+ CFG | | | | | | | | | CFG | | | | | | | | | FCG | | | | | | | | | System Call | | | | | | | | Dataset取法,取Marai(Angr),malware / benign各取一半 論文章節: 1. Introduction 2. Model Overview 描述dataset的取法,子揚取的17個Features當作Model Train的baseline(歐陽)以及related work(如果來得及再補充) 3. Evaluation 描述各種模型受到攻擊後的實際結果。 4. Conclusion 評論本次攻擊的實際情形,並評論哪一種演算法(Detector)的Robustness相對較佳。
×
Sign in
Email
Password
Forgot password
or
By clicking below, you agree to our
terms of service
.
Sign in via Facebook
Sign in via Twitter
Sign in via GitHub
Sign in via Dropbox
Sign in with Wallet
Wallet (
)
Connect another wallet
New to HackMD?
Sign up