ART 1.14.0

[beat-buesser]

This release of ART 1.14.0 introduces new privacy metrics, poisoning attacks on object detectors, the Auto Conjugate Gradient Attack, and more.

Added

• Added implementation of SHAPr membership privacy risk metric (New privacy metric shapr #1978)
• Added support for categorical non-numeric as well as continuous features in attribute inference attacks and improvements in shadow model tools (Attribute inference attack and shadow model improvements #2006)
• Added implementation of Auto Conjugate Gradient Attack for white-box evasion (Implementation of Auto Conjugate Gradient Attack #2028)
• Added implementation of adversarial training with interval bound propagation (Implement Interval Bound Propagation Training #2044)
• Added implementation of method fit to object detection estimators PyTorchFasterRCNN, PyTorchObjectDetector, and PyTorchYolo (PyTorch Object Detector Training Implementation and Prediction Update #2067)
• Added BadDet object detection poisoning attacks (RMA, GMA, OGA, ODA) (BadDet Regional Misclassification Attack Implementation #2054, BadDet Remainder Attacks Implementation #2069)

Changed

• Changed evasion detectors module by refactoring the entire module and introducing common API with the EvasionDetector base class (Refactor Evasion Detectors #1993)
• Changed loading of audio triggers with audio_perturbations to cache trigger to accelerate loading (Cache audio triggers #2053)
• Changed tested and officially supported Python versions to 3.9, 3.10, 3.11 (Update supported and tested Python versions #2063)
• Changed checks and internal improvements to AdversarialTrainerCertifiedPytorch (Zonotope updates 1.14 #2070)

Removed

[None]

Fixed

• Fixed bug in add_single_bd and add_pattern_bd to avoid confusing height and width of the trigger image and transposing the trigger (Fix Incorrect Height and Width for Image Perturbations #2046)

---

This discussion was created from the release ART 1.14.0.