Algorithm 1 Local Model Training Algorithm

1: Input: Local dataset $D_i=(X_i,{\widehat{y}}_i)$, Neural Network architecture, number of epochs, batch size   2: Output: Trained local model ${\mathcal{M}}_i$   3: procedure Local Model Training($D_i$, Neural Network, epochs, batch size)   4:    Initialize local model: ${\mathcal{M}}_i$   5:    for each epoch in 1 to $epochs$ do   6:        Shuffle $D_i$ randomly   7:        for each batch in $D_i$ with batch size $batch\_size$ do   8:            Extract batch data: $X_{batch}$, ${\widehat{y}}_{batch}$   9:            Compute predictions: ${\widehat{y}}_{pred}={\mathcal{M}}_i\left(X_{batch}\right)$ 10:           Calculate loss: $L_{batch}=Loss({\widehat{y}}_{batch},{\widehat{y}}_{pred})$ 11:           Update model weights: ${\mathcal{M}}_i.update\_weights\left(L_{batch}\right)$ 12:        end for 13:    end for 14: end procedure 15: Return Trained local model ${\mathcal{M}}_i$