Multiple calls to generate_np ignore differences in kwargs
goodfeli opened this issue · comments
Ian Goodfellow commented
Quoting Nicolas Carlini:
attack = FastGradientMethod(model, sess)
adv_1 = attack.generate_np(test_data, eps=.5)
adv_2 = attack.generate_np(test_data, eps=.2)
will result in adv_1 == adv_2, a rather unexpected result.
This is because generate_np
just stores one TensorFlow graph. It needs to have something like a dictionary mapping from argument values to graphs.
Ian Goodfellow commented
Alternately, generate_np
could feed all the values of kwargs