In your paper, you mentioned different kinds of methods for generating event space. But in the implementation of algorithms, it seems you already know which event is going to violate the differential privacy.
Take histogram for example, your implementation of the algorithm is:
def histogram(queries, epsilon): noisy_array = np.asarray(queries, dtype=np.float64) + np.random.laplace(scale=1.0 / epsilon, size=len(queries)) return noisy_array[0]
How do you know in advance that the quantization of the first element of the noisy vector would cause violation of differential privacy?
Thank you!
In your paper, you mentioned different kinds of methods for generating event space. But in the implementation of algorithms, it seems you already know which event is going to violate the differential privacy.
Take histogram for example, your implementation of the algorithm is:
def histogram(queries, epsilon): noisy_array = np.asarray(queries, dtype=np.float64) + np.random.laplace(scale=1.0 / epsilon, size=len(queries)) return noisy_array[0]How do you know in advance that the quantization of the first element of the noisy vector would cause violation of differential privacy?
Thank you!