Changed the prior vulnerability for attribute inference attack. Now, instead of considering an adversary that guesses only a random value from the sensitive attribute domain, she knows the distribution of the sensitive attribute in the dataset and guess the most frequent one. The examples and definitions were already considering this adversary.