GradientBoundednessProof<T>
Models & Types · AiDotNet.FederatedLearning.Verification
Proves that each gradient component is within [-B, B] (element-wise range proof).
Inheritance: FederatedLearningComponentBase<T> → GradientBoundednessProof<T>
Implements: IVerifiableComputation
For Beginners
While norm-bound proofs constrain the overall magnitude of a gradient, element-wise bound proofs ensure no single gradient component is abnormally large. This is important because a gradient with a small norm could still have one extreme component (a “needle” attack).
How It Works
How it works:
Optimization: Rather than proving each element individually (expensive), we use a batched proof that commits to all elements at once and proves the bound in aggregate.
Constructors
| Constructor | Summary |
|---|---|
GradientBoundednessProof(IZkProofSystem proofSystem, double elementBound) |
Initializes a new instance of GradientBoundednessProof. |
Properties
| Property | Summary |
|---|---|
string ProofSystemName { get; } |
Methods
| Method | Summary |
|---|---|
VerificationProof GenerateBoundednessProof(Tensor<T> gradient) |
Generates a proof that all gradient elements are within [-B, B]. |
VerificationProof GenerateProof(byte[] value, VerificationConstraint constraint) |
|
bool Verify(VerificationProof proof, VerificationConstraint constraint) |