ITeeSecureAggregator<T>
Interfaces · AiDotNet.FederatedLearning.TEE
Performs model aggregation inside a TEE enclave boundary.
Known implementations: TeeSecureAggregation<T>
For Beginners
In standard federated learning, the server aggregates client model updates in plaintext — the server can see every client’s update. TEE-based aggregation runs inside a secure enclave: clients encrypt their updates so only the enclave can decrypt them, the enclave aggregates in plaintext internally, and returns the sealed result. The host OS never sees individual updates.
How It Works
Flow:
Performance: 10-100x faster than homomorphic encryption because aggregation runs in plaintext inside the enclave — the overhead is only encryption/decryption at ingress/egress.
Properties
| Property | Summary |
|---|---|
bool AllUpdatesReceived { get; } |
Gets a value indicating whether all expected clients have submitted their updates. |
int UpdatesReceived { get; } |
Gets the number of updates received so far this round. |
Methods
| Method | Summary |
|---|---|
Tensor<T> Aggregate() |
Performs weighted aggregation of all submitted updates inside the TEE enclave. |
byte[] GenerateSessionKey() |
Generates a session key that clients use to encrypt their updates for this round. |
byte[] GetAttestationQuote(byte[] roundData) |
Gets the attestation quote for this aggregation enclave so clients can verify it. |
void BeginRound(int roundNumber, int expectedClients) |
Initializes the aggregator for a new round. |
void SubmitEncryptedUpdate(int clientId, byte[] encryptedUpdate, double weight) |
Submits an encrypted client model update to the enclave for aggregation. |