tf_privacy.tree_aggregation.GaussianNoiseGenerator

Gaussian noise generator with counter as pseudo state.

tf_privacy.tree_aggregation.GaussianNoiseGenerator(
    noise_std: float,
    specs: Collection[tf.TensorSpec],
    seed: Optional[int] = None
)

Produces i.i.d. spherical Gaussian noise at each step shaped according to a nested structure of tf.TensorSpecs.

Args
`noise_std`	The standard deviation of the noise.
`specs`	A nested structure of `tf.TensorSpec`s specifying the shape of the noise to generate.
`seed`	An optional integer seed. If None, generator is seeded from the clock.

Methods

initialize()

Makes an initial state for the GaussianNoiseGenerator.

Returns
A named tuple of (seeds, stddev).

make_state(
    seeds: tf.Tensor, stddev: tf.Tensor
)

Returns a new named tuple of (seeds, stddev).

next(
    state
)

Gets next value and advances the GaussianNoiseGenerator.

Args
`state`	The current state (seed, noise_std).

Returns
A tuple of (sample, new_state) where sample is a new sample and new_state is the advanced state (seed+1, noise_std).