Table 2 Hyperparameter settings for federated incremental learning (FIL) experiments.

Momentum

0.9

Update step size rate to accelerate gradient descent.

Local Epochs (E)

5

Number of training passes over local data per round.

Batch size (B)

64

Number of samples per training batch.

Initial learning rate (\(\eta\))

0.001

Learning rate used for all datasets at the start.

Training rounds

500

Total number of rounds for training across all datasets.

Regularization parameter (\(\lambda\))

Initialized to 1; optimal value via grid search

Controls overfitting, tuned through grid search.