登录
洞察 - Adversarial Attacks on Machine Learning Models
Comprehensive Evaluation of Gradient-based Adversarial Attacks: AttackBench Framework
AttackBench is a unified framework that enables a fair and comprehensive evaluation of gradient-based adversarial attacks against machine learning models. It provides a categorization of existing attacks, introduces a novel optimality metric to rank their performance, and highlights implementation issues that prevent many attacks from finding better solutions.
Manipulating Feature Importance to Generate Effective Evasion Attacks on Machine Learning Models
A novel methodology that combines SHAP-based feature importance analysis with an optimal epsilon technique to generate highly effective and precise adversarial samples that can evade machine learning models.
