Differential Privacy
class DpStep
Data for differentially private step.
Arguments:
noise_multiplier
- The ratio of the standard deviation of the Gaussian noise to the L2-sensitivity of the gradients to which the noise is added (How much noise to add).max_grad_norm
- The maximum norm of the per-sample gradients. Any gradient with norm higher than this will be clipped to this value, thus limiting the L2-sensitivity.max_batch_size
- Maximum size of the physical batch processed during computations. It will not change the size of the logical batch. If <= 0, no cap is imposed on the physical batch. Notice that due to Poisson sampling, the logical batch size during differentially private training is distributed according to a binomial distribution.
class DpBudget
Differential privacy (DP) budget.
Arguments:
eps
- Epsilon component of the DP budget.delta
- Delta component of the DP budget.