Algorithm 4: Semi-decentralized FL cluster training.

Input: $W_G$: Global Model Weights,   $NumberOfEpochs$: Number of Local Epochs   Output: $W_j$: Cluster Model Weights 1 Receive $ClusterModel$ from the server 2 Receive initial weights and parameters from the server 3 Send $ClusterModel$, weights, and parameters to the clients 4 $W_i$ += Get model weights from clients   // $\phantom{(}$ Average the received client weights within the same cluster 5  $\phantom{(}$ $W_j\leftarrow \mathrm{Average}\left(W_i\right)$ 6 return $W_j$