Adversarially Regularized Autoencoders
This work addresses the problem of representation learning for discrete data in machine learning, offering a flexible approach for tasks like text generation and style transfer, though it builds on existing frameworks like Wasserstein autoencoders.
The authors tackled the challenge of applying deep latent variable models to discrete structures like text sequences by proposing an adversarially regularized autoencoder (ARAE) method, which enables natural text generation and unaligned textual style transfer with improvements in automatic and human evaluations.
Deep latent variable models, trained using variational autoencoders or generative adversarial networks, are now a key technique for representation learning of continuous structures. However, applying similar methods to discrete structures, such as text sequences or discretized images, has proven to be more challenging. In this work, we propose a flexible method for training deep latent variable models of discrete structures. Our approach is based on the recently-proposed Wasserstein autoencoder (WAE) which formalizes the adversarial autoencoder (AAE) as an optimal transport problem. We first extend this framework to model discrete sequences, and then further explore different learned priors targeting a controllable representation. This adversarially regularized autoencoder (ARAE) allows us to generate natural textual outputs as well as perform manipulations in the latent space to induce change in the output space. Finally we show that the latent representation can be trained to perform unaligned textual style transfer, giving improvements both in automatic/human evaluation compared to existing methods.