adversarial-robustness-toolbox: ART does not work with Keras Embedding layers

Describe the bug I am unable to create any instances of art.classifiers.KerasClassifier whenever the underlying Keras model contains an Embedding layer. Using the TensorFlow backend, this invariably leads to a TypeError: Can not convert a NoneType into a Tensor or Operation.

To Reproduce Steps to reproduce the behavior:

1. Create any Keras model with an Embedding layer on the TensorFlow backend.
2. Attempt to instantiate art.classifiers.KerasClassifier on it.
3. Watch it fail.

Expected behavior I expected ART to simply return an instance of KerasClassifier as it usually does.

Screenshots N/A, but here’s a minimal non-working example:

from keras.layers import Dense, Activation, Dropout, Embedding, LSTM
from art.classifiers import KerasClassifier

model = Sequential()
model.add(Embedding(100, 128, input_length=50))
model.add(LSTM(128))
model.add(Dropout(0.5))
model.add(Dense(1))
model.add(Activation('sigmoid'))

model.compile(loss='binary_crossentropy', optimizer='rmsprop', metrics=['binary_accuracy'])

classifier = KerasClassifier((0, 1), model=model)
classifier.fit(x_train, y_train, nb_epochs=10, batch_size=128)

System information (please complete the following information):

• Ubuntu 18.04 LTS
• Python version 3.6.5
• ART version 0.5.0
• TensorFlow version 1.12.0
• Keras version 2.2.4