Learning Machine Learning Part 3: Attacking Black Box Models

In the first post in this series we covered a brief background on machine learning, the Revoke-Obfuscation approach for detecting obfuscated PowerShell scripts, and my efforts to improve the dataset and models for detecting obfuscated PowerShell. We ended up with three models: a L2 (Ridge) regularized Logistic Regression, a LightGBM Classifier, and a Neural Network architecture.

The second post covered attacking these models from a white box perspective, i.e., where we have the entirety of the trained model itself including …

adversarial-ml analytics & intelligence black box box evasion information security machine machine learning powershell