ControlVAE: Controllable Variational Autoencoder

Abstract

Variational Autoencoders (VAE) and their variants have been widely used in a variety of applications, such as dialog generation, image generation and disentangled representation learning. However, the existing VAE models have some limitations in different applications. For example, a VAE easily suffers from KL vanishing in language modeling and low reconstruction quality for disentangling. To address these issues, we propose a novel controllable variational autoencoder framework, ControlVAE, that combines a controller, inspired by automatic control theory, with the basic VAE to improve the performance of resulting generative models. Specifically, we design a new non-linear PI controller, a variant of the proportional-integral-derivative (PID) control, to automatically tune the hyperparameter (weight) added in the VAE objective using the output KL-divergence as feedback during model training. The framework is evaluated using three applications; namely, language modeling, disentangled representation learning, and image generation. The results show that ControlVAE can achieve better disentangling and reconstruction quality than the existing methods. For language modelling, it not only averts the KL-vanishing, but also improves the diversity of generated text. Finally, we also demonstrate that ControlVAE improves the reconstruction quality of generated images compared to the original VAE.

arXiv commentaccepted by ICML2020

References in corpus (15)

• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
• Stochastic Backpropagation and Approximate Inference in Deep Generative Models
• Disentangling by Factorising
• Unsupervised Image-to-Image Translation Networks
• Cyclical Annealing Schedule: A Simple Approach to Mitigating KL Vanishing
• Learning Discourse-level Diversity for Neural Dialog Models using Conditional Variational Autoencoders
• Towards Deeper Understanding of Variational Autoencoding Models
• From Variational to Deterministic Autoencoders
• Diagnosing and Enhancing VAE Models
• Variational Autoencoder with Arbitrary Conditioning
• DP-GAN: Diversity-Promoting Generative Adversarial Network for Generating Informative and Diversified Text
• Topic-Guided Variational Autoencoders for Text Generation
• Fixing a Broken ELBO
• Relevance Factor VAE: Learning and Identifying Disentangled Factors
• Generative Latent Flow

Cited by in corpus (9)

• Unifying the design space and optimizing linear and nonlinear truss metamaterials by generative modeling
• Dive into Deep Learning
• Unsupervised Belief Representation Learning with Information-Theoretic Variational Graph Auto-Encoders
• Structured Variational Inference for Simulating Populations of Radio Galaxies
• APo-VAE: Text Generation in Hyperbolic Space
• Controlling Neural Networks with Rule Representations
• Multi-type Disentanglement without Adversarial Training
• Controllable Dual Skew Divergence Loss for Neural Machine Translation
• Adaptive Compression of the Latent Space in Variational Autoencoders