How AI describes SecML-Torch

SecML-Torch is an open-source Python library for robustness evaluation and adversarial machine learning research, offering PyTorch-native support to generate adversarial examples and assess deep learning model robustness. It provides efficient native implementations of common evasion and poisoning attacks (e.g., PGD, FMN, DDN), a modular design for building custom attacks, and wrappers to external libraries like Foolbox and the Adversarial Library for broad attack coverage. It includes robustness analytics such as metrics, logging, trackers, TensorBoard integration, and tutorials, and is installable via pip with optional extras for expanded capabilities.