Added parameter checks for use_probability. Added nbviewer links to the readme.

Signed-off-by: Taesung Lee <[email protected]>

Author: TS-Lee

art/attacks/extraction/copycat_cnn.py

@@ -160,3 +160,7 @@ def _check_params(self) -> None:
 
         if not isinstance(self.nb_stolen, (int, np.int)) or self.nb_stolen <= 0:
             raise ValueError("The number of queries submitted to the victim classifier must be a positive integer.")
+
+        if not isinstance(self.use_probability, bool):
+            raise ValueError("The argument `use_probability` has to be of type bool.")
+

art/attacks/extraction/knockoff_nets.py

@@ -407,3 +407,7 @@ def _check_params(self) -> None:
 
         if not isinstance(self.verbose, bool):
             raise ValueError("The argument `verbose` has to be of type bool.")
+        if not isinstance(self.use_probability, bool):
+            raise ValueError("The argument `use_probability` has to be of type bool.")
+
+

notebooks/README.md

@@ -159,7 +159,7 @@ architecture in Keras for the CIFAR-10 dataset.
 
 ## Model stealing / model theft / model extraction
 
-[model-stealing-demo.ipynb](model-stealing-demo.ipynb) demonstrates model stealing attacks and a reverse sigmoid defense against them.
+[model-stealing-demo.ipynb](model-stealing-demo.ipynb) [[on nbviewer](https://nbviewer.jupyter.org/github/Trusted-AI/adversarial-robustness-toolbox/blob/main/notebooks/model-stealing-demo.ipynb)] demonstrates model stealing attacks and a reverse sigmoid defense against them.
 
 ## Poisoning