jax_privacy.accounting.dpsgd_event

jax_privacy.accounting.dpsgd_event(noise_multiplier, iterations, *, sampling_prob, use_zcdp=False)[source]

Returns the DpEvent for DP-SGD with the given training parameters.

This mechanism is a composition of poisson-sampled Gaussian mechanisms. See this paper for more details:

Parameters:
  • noise_multiplier (float) – The noise multiplier of the mechanism.

  • iterations (int) – The number of iterations to run the mechanism for.

  • sampling_prob (float) – The Poisson sampling probability of the mechanism, i.e., the probability an example will be included in each batch.

  • use_zcdp (bool) – Whether to use generic dp_event.ZCDpEvent instead of specific dp_event.GaussianDpEvent. For example, set this to True if using discrtete Gaussian noise instead of continuous Gaussian noise.

Return type:

DpEvent

Returns:

A DpEvent object.