Algorithm 1: Reselecting Data Progressively.

Parameters: M = 251: total epoch number for training, E = 50: number of epochs included in one stage of reselecting data progressively, D: all the training data, R = 4: ratio of overall data selection, List = [25, 10, 5, 2]: a list of epoch values for reselecting data, S = [0, List[0]]: a list for saving the number of rounds for which data should be reselected, L = [0, …, len(D)]: a list of integers from 0 to the length of D, Loader: dataloader in Pytorch

1. for i = 0 to len(List) do 2.       while S[−1] < (E*(i + 2)) do 3.             S.append(List[i] + S[−1]) 4.       end while 5. end for 6. for i = 0 to M do 7.       if i in S then 8.              Shuffle(L) 9.              Part = [D[j] for j in L[0:(len(D)//R)]] 10.             Loader(Part) 11.     end if 12.     Train one epoch 13. end for