一种基于对抗MSE度量的自编码网络优化方法

An auto-encoder network optimization method based on adversarial Mean Square Error

自编码网络是一种常用的无监督图像特征提取和压缩方法,通常使用均方差(Mean Square Error,MSE)作为损失函数。然而,由于MSE没有考虑图像的分布特性,导致自编码网络性能下降。提出一种新的对抗MSE度量方法,在自编码网络的基础上增加权重网络,训练自编码网络的同时对抗训练权重网络,基于对抗网络生成加权均方差的权重,将图像的分布特性引入损失函数,进而提升自编码网络的特征提取和重构能力。

The auto-encoder network is a commonly used unsupervised image feature extraction and data compression method,which usually takes Mean Square Error(MSE)as the loss function.However,as MSE does not take the distribution characteristics of the image into account,the performance of the auto-encoder network is degraded.To solve this problem,a new algorithm called adversarial MSE is proposed by adding weight network to the auto-encoder network and training them simultaneously.The weighted MSE is generated based on the adversarial network,and the distribution characteristics of the image are introduced into the loss function,so as to improve the feature extraction and reconstruction capabilities of the auto-encoder network.