The train file in this directory will train an ANE NOTA model, Adversarial NOTA Envelopment model for either cifar10, cifar100 or Tiny Imagenet.

Our trained ANE, Boundry Padding, adversarial training and undefended models are provided.  

The Test file is capable of deterministic or stochastic testing. This paper only uses deterministic testing. A following paper is investigating Bayesian versions of these defenses. As long as the adapted attack files are placed in the evasion folder of the art library (v1.11) in the users environment, and the test code will call the specified attacks, with the variables passed at the command line.  
Passing the NOTA variable as a bool of true or false for whether the underlying model is NOTA defended or not. The assumption is that an additonal class is moved on to the end, so that the last class is the NOTA class.

 
