Discriminative Regularization for Generative Models
This work addresses the challenge of enhancing sample quality in generative models for applications like image and speech generation, though it is incremental as it builds on existing variational autoencoder methods.
The authors tackled the problem of improving generative model quality by using representations from discriminative classifiers to augment the objective function of variational autoencoders, resulting in samples that are clearer and have higher visual quality than standard models.
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.