GradientNormRangeProof<T>
Models & Types · AiDotNet.FederatedLearning.Verification
Generates and verifies proofs that a gradient’s L2 norm is within a declared bound.
Inheritance: FederatedLearningComponentBase<T> → GradientNormRangeProof<T>
Implements: IVerifiableComputation
For Beginners
Gradient clipping ensures no single client can have outsized influence on the global model. But how does the server verify that the client actually clipped its gradient? This class generates a cryptographic proof that ||g|| <= C without revealing the actual norm value.
How It Works
How it works (simplified Bulletproofs approach):
Reference: Bulletproofs (Bunz et al., S&P 2018) for logarithmic-size range proofs.
Constructors
| Constructor | Summary |
|---|---|
GradientNormRangeProof(IZkProofSystem proofSystem, double normBound) |
Initializes a new instance of GradientNormRangeProof. |
Properties
| Property | Summary |
|---|---|
string ProofSystemName { get; } |
Methods
| Method | Summary |
|---|---|
VerificationProof GenerateNormProof(Tensor<T> gradient) |
Generates a proof that the gradient’s L2 norm is within [0, C]. |
VerificationProof GenerateProof(byte[] value, VerificationConstraint constraint) |
|
bool Verify(VerificationProof proof, VerificationConstraint constraint) |