Skip to content

Differential privacy

DpBudget dataclass 

DpBudget(
    eps: PositiveFloat,
    delta: Annotated[float, Field(ge=0, lt=1)],
)

Differential privacy (DP) budget.

Parameters:

Name Type Description Default
eps PositiveFloat

Epsilon component of the DP budget.

 required
delta Annotated[float, Field(ge=0, lt=1)]

Delta component of the DP budget.

 required

DpStep pydantic-config 

DpStep(
    noise_multiplier: PositiveFloat = 1.0,
    max_grad_norm: PositiveFloat = 1.0,
    max_batch_size: NonNegativeInt = 0,
)

Data for differentially private step.

Parameters:

Name Type Description Default
noise_multiplier PositiveFloat

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).

 1.0
max_grad_norm PositiveFloat

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.

 1.0
max_batch_size NonNegativeInt

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.

 0