Multi-Adversarial Autoencoders: Stable, Faster and Self-Adaptive Representation Learning

The variational autoencoder (VAE) and generative adversarial networks (GAN) are two prominent approaches to achieving a probabilistic generative model by way of an autoencoder and a two-player minimax game. While VAEs often suffer from over-simplified posterior approximations, the adversarial autoencoder (AAE) has shown promise by adopting GAN to match the variational posterior to an arbitrary prior through adversarial training. In this paper, we propose the Multi-adversarial Autoencoder (MAAE), which extends the AAE framework by incorporating multiple discriminators and enabling softensemble feedback. By adaptively regulating the collective feedback from multiple discriminators, MAAE captures a balance between fitting the data distribution and performing correct inference and accelerates training stability while extracting meaningful and interpretable latent representations. Experimental evaluations on MNIST, CIFAR10, and CelebA datasets demonstrate significant improvements in latent representation, quality of generated samples, loglikelihood, and a pairwise comparison metric.

Recommended citation: Wu, Xinyu and Jang, Hyeryung, Multi-Adversarial Autoencoders: Stable, Faster and Self-Adaptive Representation Learning. Available at SSRN: https://ssrn.com/abstract=4530466 or http://dx.doi.org/10.2139/ssrn.4530466
Download Paper