How AI describes RobustCheck

RobustCheck is a Python package for evaluating the robustness of image classification models by applying black-box untargeted adversarial attacks and computing robustness metrics. It targets models that expose a batch_predict function returning probability distributions and includes implementations of EvoBA and Epsilon-Greedy attacks, with tooling to run robustness checks, save results, and generate MLflow logs. The package is designed to be extensible, featuring abstract classes for UntargetedAttack and EvoStrategy to encourage adding new attacks and strategies.