tf_privacy.DistributedSkellamSumQuery

Implements DPQuery interface for discrete distributed sum queries.

Inherits From: SumAggregationDPQuery, DPQuery

This implementation is for the distributed queries where the Skellam noise is applied locally to a discrete vector that matches the norm bound.

l1_norm_bound The l1 norm bound to verify for each record.
l2_norm_bound The l2 norm bound to verify for each record.
local_stddev The standard deviation of the Skellam distribution.

Methods

accumulate_preprocessed_record

View source

Implements tensorflow_privacy.DPQuery.accumulate_preprocessed_record.

accumulate_record

View source

Accumulates a single record into the sample state.

This is a helper method that simply delegates to preprocess_record and accumulate_preprocessed_record for the common case when both of those functions run on a single device. Typically this will be a simple sum.

Args
params The parameters for the sample. In standard DP-SGD training, the clipping norm for the sample's microbatch gradients (i.e., a maximum norm magnitude to which each gradient is clipped)
sample_state The current sample state. In standard DP-SGD training, the accumulated sum of previous clipped microbatch gradients.
record The record to accumulate. In standard DP-SGD training, the gradient computed for the examples in one microbatch, which may be the gradient for just one example (for size 1 microbatches).

Returns
The updated sample state. In standard DP-SGD training, the set of previous microbatch gradients with the addition of the record argument.

add_noise_to_sample

View source

Adds Skellam noise to the sample.

We use difference of two Poisson random variable with lambda hyperparameter that equals 'local_stddev**2/2' that results in a standard deviation 'local_stddev' for the Skellam noise to be added locally.

Args
local_stddev The standard deviation of the local Skellam noise.
record The record to be processed.

Returns
A record with added noise.

derive_metrics

View source

Derives metric information from the current global state.

Any metrics returned should be derived only from privatized quantities.

Args
global_state The global state from which to derive metrics.

Returns
A collections.OrderedDict mapping string metric names to tensor values.

derive_sample_params

View source

Given the global state, derives parameters to use for the next sample.

For example, if the mechanism needs to clip records to bound the norm, the clipping norm should be part of the sample params. In a distributed context, this is the part of the state that would be sent to the workers so they can process records.

Args
global_state The current global state.

Returns
Parameters to use to process records in the next sample.

get_noised_result

View source

The noise was already added locally, therefore just continue.

initial_global_state

View source

Since we operate on discrete values, use int for L1 bound and float for L2 bound.

initial_sample_state

View source

Implements tensorflow_privacy.DPQuery.initial_sample_state.

merge_sample_states

View source

Implements tensorflow_privacy.DPQuery.merge_sample_states.

preprocess_record

View source

Check record norm and add noise to the record.

For both L1 and L2 norms we compute a global norm of the provided record. Since the record contains int32 tensors we cast them into float32 to compute L2 norm. In the end we run three asserts: type, l1, and l2 norms.

Args
params The parameters for the particular sample.
record The record to be processed.

Returns
A tuple (preprocessed_records, params) where preprocessed_records is the structure of preprocessed tensors, and params contains sample params.