ART 1.10.2 #1711
beat-buesser
announced in
Announcements
ART 1.10.2
#1711
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
This release of ART 1.10.2 provides updates to ART 1.10.
Added
[None]
Changed
PyTorchClassifierto use a new optimizer when cloned withclone_for_refitting(PyTorchClassifier: use a new optimizer for the cloned classifier #1580)art.estimators.gan.*andart.estimators.generator.*to follow naming convention (Update DGM poisoning attacks for TensorFlow dependency #1655)Mp3CompressionPyTorchandPyTorchDeepSpeechto add support for samples in 2D non-object arrays (Mp3Compression defense bug fix - handle case when input is 2D but not an object array #1680, Add support for framework-specific preprocessing of object arrays #1702)python_object_detector.pytopytorch_object_detector.pyto follow naming convention (Rename python_object_detector.py to pytorch_object_detector.py #1687)CarliniLInfMethodby adding argument forbatch_size(Add argument batch_size to CarliniLInfMethod #1699).Removed
[None]
Fixed
ImperceptibleASRPyTorchby adding missing.detach().cpu()and.cpu()calls (Imperceptible asr bug detach #1677)art.estimators.certification.randomized_smoothingestimators to correctly apply Gaussian noise (Randomised smoothing fix #1678)GaussianNoisethe post-processing defence to keep number of dimensions constant during normalisation (fix: fixed np.sum in art/defences/postprocessor/gaussian_noise.py #1684)RobustDPatchfor channels first images to correctly un-transform loss gradients (Update RobustDPatch for channels first images #1693)PoisoningAttackCleanLabelBackdoor(Support numpy arrays in logger of clean_label_backdoor_attack #1698)This discussion was created from the release Bump version to ART 1.10.2.
Beta Was this translation helpful? Give feedback.
All reactions