Discriminative Regularization for Generative Models
Alex Lamb, Vincent Dumoulin and Aaron Courville Montreal Institute for Learning Algorithms, Université de Montréal
Abstract
We explore the question of whether the representations learned by classifiers can be used to enhance the quality of generative models. Our conjecture is that labels correspond to characteristics of natural data which are most salient to humans: identity in faces, objects in images, and utterances in speech. We propose to take advantage of this by using the representations from discriminative classifiers to augment the objective function corresponding to a generative model. In particular we enhance the objective function of the variational autoencoder, a popular generative model, with a discriminative regularization term. We show that enhancing the objective function in this way leads to samples that are clearer and have higher visual quality than the samples from the standard variational autoencoders.
中文速览
生成式模型（如变分自编码器VAE）产生的图像往往模糊、缺乏语义细节，根本原因在于训练时用的像素级L²损失与人类视觉感知严重不符。作者提出"判别式正则化"方法：在VAE的标准训练目标之外，额外要求模型重建出的图像在一个预训练分类器（卷积神经网络）各隐藏层的特征表示空间中也尽量贴近真实图像，从而让低层特征匹配保证边缘清晰、高层特征匹配保证语义一致。在SVHN、CIFAR-10和CelebA三个基准数据集上的实验表明，这种方法生成的图像比标准VAE明显更清晰、视觉质量更高。这项工作的意义在于为生成模型提供了一种简洁实用的改进思路——借用判别模型已经学到的丰富视觉语义表示来弥补生成模型的感知短板，无需改变整体架构。
原文 arXiv:1602.03220;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1602.03220v4