Stein Latent Optimization for Generative Adversarial Networks

TL;DR

SLOGAN introduces Stein latent optimization for GANs, enabling unsupervised learning of balanced or imbalanced attributes in latent spaces, with effective attribute manipulation using minimal probe data.

Contribution

It proposes a novel Stein latent optimization method with an encoder and contrastive loss for improved unsupervised attribute learning in GANs.

Findings

Successfully learns balanced and imbalanced attributes

Achieves state-of-the-art unsupervised conditional generation

Allows attribute manipulation with minimal probe data

Abstract

Generative adversarial networks (GANs) with clustered latent spaces can perform conditional generation in a completely unsupervised manner. In the real world, the salient attributes of unlabeled data can be imbalanced. However, most of existing unsupervised conditional GANs cannot cluster attributes of these data in their latent spaces properly because they assume uniform distributions of the attributes. To address this problem, we theoretically derive Stein latent optimization that provides reparameterizable gradient estimations of the latent distribution parameters assuming a Gaussian mixture prior in a continuous latent space. Structurally, we introduce an encoder network and novel unsupervised conditional contrastive loss to ensure that data generated from a single mixture component represent a single attribute. We confirm that the proposed method, named Stein Latent Optimization for GANs (SLOGAN), successfully learns balanced or imbalanced attributes and achieves state-of-the-art unsupervised conditional generation performance even in the absence of attribute information (e.g., the imbalance ratio). Moreover, we demonstrate that the attributes to be learned can be manipulated using a small amount of probe data.