UnlearningCertificate
Models & Types · AiDotNet.FederatedLearning.Unlearning
Certificate proving that a client’s data has been unlearned from the federated model.
For Beginners
When a client requests data removal, the system needs to prove it actually happened. This certificate contains verifiable metrics showing the client’s contribution has been removed. Think of it as a “receipt of forgetting” — proof for GDPR auditors that the right to be forgotten was honored.
How It Works
Key metrics:
Constructors
| Constructor | Summary |
|---|---|
UnlearningCertificate() |
Properties
| Property | Summary |
|---|---|
DateTime Timestamp { get; set; } |
Gets or sets the UTC timestamp when unlearning was performed. |
UnlearningMethod MethodUsed { get; set; } |
Gets or sets the unlearning method used. |
bool Verified { get; set; } |
Gets or sets whether the unlearning was verified as correct. |
double MembershipInferenceScore { get; set; } |
Gets or sets the membership inference attack score after unlearning. |
double ModelDivergence { get; set; } |
Gets or sets the L2 distance between the original and unlearned model parameters. |
double RetainedAccuracy { get; set; } |
Gets or sets the retained accuracy on non-target clients after unlearning. |
int ClientRoundsParticipated { get; set; } |
Gets or sets the number of training rounds the target client participated in. |
int TargetClientId { get; set; } |
Gets or sets the client ID whose data was unlearned. |
long UnlearningTimeMs { get; set; } |
Gets or sets the wall-clock time in milliseconds the unlearning took. |
string PostUnlearningModelHash { get; set; } |
Gets or sets a hash of the model state after unlearning. |
string PreUnlearningModelHash { get; set; } |
Gets or sets a hash of the model state before unlearning (for audit trail). |
string Summary { get; set; } |
Gets or sets an optional human-readable summary of the unlearning result. |