Are Keras discarded and the GaussianNoise layers apply different cases of noise in each batch?

Question

I am creating an auto noise reduction in Keras. The model I'm using is

input_img = Input(shape=(10,)) encoded = GaussianNoise(0.01)(input_img) encoded = Dropout(0.1)(encoded) encoded = Dense(20,activation='relu')(encoded) decoded = Dense(10, activation='sigmoid')(encoded) ae = Model(input=input_img, output=decoded) 

If I call later

 ae.fit(x_train, x_train, nb_epoch=3, batch_size=5, shuffle=True, validation_data=(x_test, x_test)) 

Is there a new instance of noise created for each batch? In other words, for each era above, are there different cases of noise for each party? Or is the noise instance fixed at the same thing for all parties and only changes when the era changes? Or worse, only one instance of noise is selected for everything?

Answer 1

Each instance of noise is created every time (that is, each batch) that the layer invocation function is used during training.