Algorithm 2. Partitioning Algorithm for Edge Server i.
1:  2:  3:  4:  5:  6:  7:	Input: Split_i (can be allocated from the parameter server)             Model_pre (pre-trained model)             Workload  Output: Local_model_i  Split_i ← getSplitFromPS();  Local_model_i ← train(Workload, Model_pre); // backpropagation  return Local_model_i;