然而，当我们只考虑基底自动编码器的输入和最终输出之间的距离时，我们经常

然而，当我们只考虑基底自动编码器的输入和最终输出之间的距离时，我们经常观察到一个过拟合问题和不稳定的训练收敛性。为了解决这些问题，我们设计了一种基于多阶段距离的新损耗，它包含了由部分自动编码器获得的输入和输出之间的距离。部分自动编码器{AEI(X)}Nli=1仅包含其原始自动编码器AE(X)的一部分编码和解码层，而输入和输出大小与原始自动编码器(X)=g1（f1(X)）（g2（f2(X)））（AE(X)(gNL)（f1(X)））（g2（f1(X)））。因此，基于多级距离的损失可以描述为：

However, when we only consider the distance between the input and final output of the base automatic encoder, we often observe an over fitting problem and unstable training convergence. In order to solve these problems, we design a new loss based on multi-stage distance, which includes the distance between input and output obtained by some automatic encoders. The partial automatic encoder {AEI (x)} NLI = 1 contains only a part of the encoding and decoding layer of its original automatic encoder AE (x), and the input and output sizes are the same as those of the original automatic encoder (x) = G1 (F1 (x)) (G2 (F2 (x))) (AE (x) (GNL) (F1 (x)) (G2 (F1 (x)). Therefore, the loss based on multi-level distance can be described as:

However, when we only consider the distance between the input and the final output of the automatic encoder, we often observe an over-fitting problem and unstable training convergence. In order to solve these problems, we designed a new loss based on multi-stage distance, which includes the distance between input and output obtained by partial automatic encoder. Partial automatic encoder {AEI(X)}Nli=1 only contains a part of coding and decoding layers of its original automatic encoder AE(X), while the input and output sizes are the same as those of the original automatic encoder (x) = G1 (F1 (x)) (G2 (F2 (x)) (AE (x) (GNL) (F1 (x)) Therefore, the loss based on multi-level distance can be described as: