Skip to content

dpgan

DPGAN

Bases: DPMixin, GAN

A differentially private GAN. Accepts GAN arguments as well as DPMixin arguments.

DP is applied to the discriminator only, as it is the sole component that processes real data.

Source code in src/nhssynth/modules/model/models/dpgan.py 
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
class DPGAN(DPMixin, GAN):
    """
    A differentially private GAN. Accepts [`GAN`][nhssynth.modules.model.models.gan.GAN] arguments
    as well as [`DPMixin`][nhssynth.modules.model.common.dp.DPMixin] arguments.

    DP is applied to the discriminator only, as it is the sole component that processes real data.
    """

    def __init__(
        self,
        *args,
        target_epsilon: float = 3.0,
        target_delta: Optional[float] = None,
        max_grad_norm: float = 5.0,
        secure_mode: bool = False,
        **kwargs,
    ) -> None:
        super(DPGAN, self).__init__(
            *args,
            target_epsilon=target_epsilon,
            target_delta=target_delta,
            max_grad_norm=max_grad_norm,
            secure_mode=secure_mode,
            **kwargs,
        )

    def make_private(self, num_epochs: int) -> GradSampleModule:
        """
        Make the [`discriminator`][nhssynth.modules.model.models.gan.GAN] differentially private.

        Only the discriminator is privatised since only it processes real training data.

        Args:
            num_epochs: The number of epochs to train for
        """
        self.discriminator = super().make_private(num_epochs, self.discriminator)

    @classmethod
    def get_args(cls) -> list[str]:
        return GAN.get_args() + DPMixin.get_args()

    @classmethod
    def get_metrics(cls) -> list[str]:
        return GAN.get_metrics() + DPMixin.get_metrics()

make_private(num_epochs)

Make the discriminator differentially private.

Only the discriminator is privatised since only it processes real training data.

Parameters:

Name Type Description Default
num_epochs int

The number of epochs to train for

 required
Source code in src/nhssynth/modules/model/models/dpgan.py 
35
36
37
38
39
40
41
42
43
44
def make_private(self, num_epochs: int) -> GradSampleModule:
    """
    Make the [`discriminator`][nhssynth.modules.model.models.gan.GAN] differentially private.

    Only the discriminator is privatised since only it processes real training data.

    Args:
        num_epochs: The number of epochs to train for
    """
    self.discriminator = super().make_private(num_epochs, self.discriminator)