IPrivacyAccountant
Interfaces · AiDotNet.Interfaces
Tracks cumulative privacy loss across federated learning rounds.
Known implementations: PrivacyAccountantBase
How It Works
For Beginners: Differential privacy has a finite “budget” (epsilon, delta). Each training round spends some of that budget. A privacy accountant keeps track of what was spent so you can report guarantees and enforce limits.
Methods
| Method | Summary |
|---|---|
double GetEpsilonAtDelta(double targetDelta) |
Gets a reported epsilon value at a given target delta (if supported by the accountant). |
double GetTotalDeltaConsumed() |
Gets the total delta consumed so far according to this accountant. |
double GetTotalEpsilonConsumed() |
Gets the total epsilon consumed so far according to this accountant. |
string GetAccountantName() |
Gets the name of this privacy accountant implementation. |
void AddRound(double epsilon, double delta, double samplingRate) |
Records a single privacy event (typically one federated learning round). |