8184b9fdf51e3f75835fe1f2d56c294d16686241,examples/federated_learning_with_encryption.py,,,#,162

Before Change

    // Predict
    for (i, c) in enumerate(clients):
        print("Train", mean_square_error(c.predict(X[i]), y[i]))
        print("Test", mean_square_error(c.predict(X_test), y_test))

    // Each client sends its own model to the next one, in a RING protocol,
    // aggregating them all. The last client sends the aggregate model to the server

After Change

    //     print(mean_square_error(y_pred, y_test))

    // The federated learning with gradient from the google paper
    n_iter = 5
    for i in range(n_iter):

        // Compute gradients, encrypt and aggregate
        encrypt_aggr = clients[0].encrypted_gradient(sum_to=None)
        encrypt_aggr = clients[1].encrypted_gradient(sum_to=encrypt_aggr)
        encrypt_aggr = clients[2].encrypted_gradient(sum_to=encrypt_aggr)

        // Send aggregate to server, which decrypts
        aggr = server.decrypt_aggregate(encrypt_aggr, n_clients)

        // Take gradient steps
        clients[0].gradient_step(aggr)
        clients[1].gradient_step(aggr)
        clients[2].gradient_step(aggr)

    for (i, c) in enumerate(clients):
        y_pred = c.predict(c.X)
        print(mean_square_error(y_pred, c.y))

Instances