TeeSecureAggregation<T>
Models & Types ยท AiDotNet.FederatedLearning.TEE
Performs weighted model aggregation inside a TEE enclave boundary.
Inheritance: FederatedLearningComponentBase<T> โ TeeSecureAggregation<T>
Implements: ITeeSecureAggregator<T>
For Beginners
This is the core component that makes TEE-based federated learning work. Instead of aggregating model updates in regular memory (where the server OS can see them), this class runs aggregation inside a hardware-protected enclave:
How It Works
Performance: TEE aggregation is 10-100x faster than homomorphic encryption because the enclave operates on plaintext internally. The only overhead is AES encryption/decryption at the enclave boundary.
Security model: The host OS and hypervisor cannot read or modify data inside the enclave. Even a compromised server cannot extract individual client updates. Only the aggregated result leaves the enclave.
Constructors
| Constructor | Summary |
|---|---|
TeeSecureAggregation(ITeeProvider<T> provider, TeeOptions options) |
Initializes a new instance of TeeSecureAggregation. |
Properties
| Property | Summary |
|---|---|
bool AllUpdatesReceived { get; } |
|
int UpdatesReceived { get; } |
Methods
| Method | Summary |
|---|---|
Tensor<T> Aggregate() |
|
byte[] EncryptForSubmission(Tensor<T> parameters) |
Encrypts model parameters for submission to this aggregator. |
byte[] GenerateSessionKey() |
|
byte[] GetAttestationQuote(byte[] roundData) |
|
void BeginRound(int roundNumber, int expectedClients) |
|
void SubmitEncryptedUpdate(int clientId, byte[] encryptedUpdate, double weight) |