Prompt

we consider a FL system, consisting of an edge server and $K$ numbers of mobile devices.Traditionally, FL involves $L$ numbers of global training rounds, indexed by $\ell \in \mathcal{L}=\{1, 2, \cdots, L\}$. In this paper, we group every $T_{\ell}$ numbers of global training rounds as one time slot, denoted as $t$. Suppose that there exist $T$ numbers of time slots, and within each time slot $t \in \mathcal{T}$, we perform pruning global model to preserve the structured sparsity and model performance. In a global training round $\ell$, we further prune parameters of the pruned global model.At the beginning of round $t$, each device uploads its up--to--date states, such as computing capacity, to its DT(digital twin). The DT will determine $\Psi_{k}^{t,\ell}$ and $\Phi_{k}^{t,\ell}$, which will be detailed later. Then, based on $\Psi_{k}^{t,\ell}$, the DT will prune the global model $\omega$ to obtain ${\omega}_{k}^{t,\ell}$, and train ${\omega}_{k}^{t,\ell}$ on its synthetic dataset $\hat{\mathcal{D}}_k$.At time slot $\ell$, device $k$ trains the received model ${\omega}_{k}^{t,\ell}$ on $\mathcal{D}_k$ for $\lambda_k$ numbers of local iterations to obtain $\tilde{{\omega}}_{k}^{t,\ell}$.Based on received $\Phi_{k}^{t,\ell}$, device $k$ prunes the model $\tilde{{\omega}}_{k}^{t,\ell}$.Once completion of local model training, devices should transmit their models to DTs. Upon the receipt of all local models, edge server implement the aggregation scheme at each slot $\ell$.At the next global training round, the edge server transfers the aggregated model to DTs, and a next learning process is launched. 注意:重点突出time slot和time round

Result

GPT Image-2 - Text to Image

Settings