IVerifiableComputation

Interfaces · AiDotNet.FederatedLearning.Verification

Defines the contract for generating and verifying proofs of computation correctness.

Known implementations: ComputationIntegrityProof<T>, GradientBoundednessProof<T>, GradientNormRangeProof<T>, LossThresholdProof<T>

For Beginners

In federated learning, the server receives gradient updates from clients. But how does the server know a client actually trained correctly? A malicious client could send random numbers or carefully crafted poisoning updates.

How It Works

Verifiable computation solves this: the client generates a cryptographic “proof” alongside its gradient that the server can check in milliseconds — even though generating the proof takes much longer. If the proof verifies, the server has mathematical certainty that the computation was done correctly.

Types of proofs:

Properties

Property Summary
string ProofSystemName { get; } Gets the name of this proof system.

Methods

Method Summary
VerificationProof GenerateProof(byte[] value, VerificationConstraint constraint) Generates a proof that a value satisfies a constraint.
bool Verify(VerificationProof proof, VerificationConstraint constraint) Verifies a proof generated by GenerateProof.