ART 1.13.0

[beat-buesser]

This release of ART 1.13.0 introduces black-box regression estimator, DP-InstaHide, object detection estimator for TensorFlow v2, and more.

Added

• Added CutOut data augmentation as preprocessor in Numpy, TensorFlow and PyTorch (Implement Cutout in Numpy, PyTorch, and TensorFlow #1850)
• Added MixUp data augmentation as preprocessor in Numpy, TensorFlow and PyTorch (Implement Mixup in Numpy, PyTorch, and TensorFlow #1885)
• Added CutMix data augmentation as preprocessor in Numpy, TensorFlow and PyTorch (Implement CutMix in Numpy, PyTorch, and TensorFlow #1910)
• Added regression estimator for black-box scenario (Blackbox regressor support + shadow support for additional models #1930)
• Added additional model support for shadow models (Blackbox regressor support + shadow support for additional models #1930)
• Added Numpy-based data generator to support very large datasets (numpy data generators #1934
• Added object detection estimator for Faster-RCNN in TensorFlow v2 (Adding TFv2 support for Faster-RCNN #1951)
• Added DP-InstaHide training for classification with differentially private data augmentations (DP-InstaHide Trainer Implementation #1956)
• Added Interval Bound Propagation for certified classification in PyTorch (Add Interval Bound Propagation for classification #1965)

Changed

[None]

Removed

[None]

Fixed

• Fixed unexpected shape in art.utils.load_cifar10 for loading raw dataset (Fix incorrect shape issue when loading raw CIFAR-10 data #1962)
• Fixed bug to return correct best poisoning indices in SleeperAgentAttack (bug fix for getting best poison indices after all trials #1955)

---

This discussion was created from the release ART 1.13.0.