LossThresholdProof<T>
Models & Types · AiDotNet.FederatedLearning.Verification
Proves that a client’s local training loss is below a threshold without revealing the actual value.
Inheritance: FederatedLearningComponentBase<T> → LossThresholdProof<T>
Implements: IVerifiableComputation
For Beginners
After local training, each client has a training loss value. If the loss is too high, the client’s model may be poorly trained or even poisoned. This proof lets the server reject clients with high loss without seeing the actual loss value (which could leak information about the client’s private dataset).
How It Works
How it works:
Reference: ZKP-FedEval (2025) — privacy-preserving FL evaluation using ZKPs.
Constructors
| Constructor | Summary |
|---|---|
LossThresholdProof(IZkProofSystem proofSystem, double threshold) |
Initializes a new instance of LossThresholdProof. |
Properties
| Property | Summary |
|---|---|
string ProofSystemName { get; } |
Methods
| Method | Summary |
|---|---|
VerificationProof GenerateLossProof(double loss, int clientId, int round) |
Generates a proof that the given loss value is below the threshold. |
VerificationProof GenerateProof(byte[] value, VerificationConstraint constraint) |
|
bool Verify(VerificationProof proof, VerificationConstraint constraint) |