On the Generative Utility of Cyclic Conditionals

Abstract

We study whether and how can we model a joint distribution $p(x,z)$ using two conditional models $p(x|z)$ and $q(z|x)$ that form a cycle. This is motivated by the observation that deep generative models, in addition to a likelihood model $p(x|z)$, often also use an inference model $q(z|x)$ for extracting representation, but they rely on a usually uninformative prior distribution $p(z)$ to define a joint distribution, which may render problems like posterior collapse and manifold mismatch. To explore the possibility to model a joint distribution using only $p(x|z)$ and $q(z|x)$, we study their compatibility and determinacy, corresponding to the existence and uniqueness of a joint distribution whose conditional distributions coincide with them. We develop a general theory for operable equivalence criteria for compatibility, and sufficient conditions for determinacy. Based on the theory, we propose a novel generative modeling framework CyGen that only uses the two cyclic conditional models. We develop methods to achieve compatibility and determinacy, and to use the conditional models to fit and generate data. With the prior constraint removed, CyGen better fits data and captures more representative features, supported by both synthetic and real-world experiments.

arXiv commentNeurIPS'21 camera-ready version

References in corpus (23)

• Adam: A Method for Stochastic Optimization
• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• Auto-Encoding Variational Bayes
• GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium
• Variational Inference with Normalizing Flows
• Learning to Discover Cross-Domain Relations with Generative Adversarial Networks
• Dual Learning for Machine Translation
• Stochastic Gradient Hamiltonian Monte Carlo
• Deep Generative Stochastic Networks Trainable by Backprop
• On the Identifiability of the Post-Nonlinear Causal Model
• Generalized Denoising Auto-Encoders as Generative Models
• Residual Flows for Invertible Generative Modeling
• Analyzing Inverse Problems with Invertible Neural Networks
• DiffWave: A Versatile Diffusion Model for Audio Synthesis
• Explorations in Homeomorphic Variational Auto-Encoding
• A Generative Process for Sampling Contractive Auto-Encoders
• A Unified Particle-Optimization Framework for Scalable Bayesian Sampling
• Understanding and Accelerating Particle-Based Variational Inference
• Coupling-based Invertible Neural Networks Are Universal Diffeomorphism Approximators
• Uncertainty Autoencoders: Learning Compressed Representations via Variational Information Maximization
• Graphical Generative Adversarial Networks
• Bidirectional Helmholtz Machines
• Quantitative Understanding of VAE as a Non-linearly Scaled Isometric Embedding