Struct DiffPrivacyEngine 

pub struct DiffPrivacyEngine { /* private fields */ }

Differential privacy engine for adding calibrated noise.

Implementations

impl DiffPrivacyEngine

pub fn gaussian( epsilon: f64, delta: f64, sensitivity: f64, clipping_norm: f64, ) -> Result<Self, FederationError>

Create a new DP engine with Gaussian mechanism.

Default: epsilon=1.0, delta=1e-5 (strong privacy).

pub fn laplace( epsilon: f64, sensitivity: f64, clipping_norm: f64, ) -> Result<Self, FederationError>

Create a new DP engine with Laplace mechanism.

pub fn with_seed(self, seed: u64) -> Self

Create with a deterministic seed (for testing).

pub fn clip_gradients(&self, gradients: &mut [f64])

Clip a gradient vector to the configured L2 norm bound.

pub fn add_noise(&mut self, params: &mut [f64]) -> DiffPrivacyProof

Add calibrated noise to a vector of parameters.

Clips gradients first, then adds noise per the configured mechanism.

pub fn add_noise_scalar(&mut self, value: &mut f64) -> f64

Add noise to a single scalar value.

pub fn epsilon(&self) -> f64

Current epsilon setting.

pub fn delta(&self) -> f64

Current delta setting.