Towards Deeper Understanding of Variational Autoencoding Models

Abstract

We propose a new family of optimization criteria for variational auto-encoding models, generalizing the standard evidence lower bound. We provide conditions under which they recover the data distribution and learn latent features, and formally show that common issues such as blurry samples and uninformative latent features arise when these conditions are not met. Based on these new insights, we propose a new sequential VAE model that can generate sharp samples on the LSUN image dataset based on pixel-wise reconstruction loss, and propose an optimization criterion that encourages unsupervised learning of informative latent features.

References in corpus (6)

• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
• Autoencoding beyond pixels using a learned similarity metric
• f-GAN: Training Generative Neural Samplers using Variational Divergence Minimization
• Improving Variational Inference with Inverse Autoregressive Flow
• PixelCNN++: Improving the PixelCNN with Discretized Logistic Mixture Likelihood and Other Modifications
• Discriminative Regularization for Generative Models

Cited by in corpus (49)

• Multimodal Generative Models for Scalable Weakly-Supervised Learning
• BIVA: A Very Deep Hierarchy of Latent Variables for Generative Modeling
• From Variational to Deterministic Autoencoders
• Deep Generative Modelling: A Comparative Review of VAEs, GANs, Normalizing Flows, Energy-Based and Autoregressive Models
• ControlVAE: Controllable Variational Autoencoder
• Variational Inference of Disentangled Latent Concepts from Unlabeled Observations
• Avoiding Latent Variable Collapse With Generative Skip Models
• Semi-Amortized Variational Autoencoders
• Autoregressive Quantile Networks for Generative Modeling
• Hierarchical Quantized Autoencoders
• Advances in Variational Inference
• Leveraging Variational Autoencoders for Parameterized MMSE Estimation
• The Riemannian Geometry of Deep Generative Models
• Diverse Image Inpainting with Bidirectional and Autoregressive Transformers
• Intrinsically Motivated Discovery of Diverse Patterns in Self-Organizing Systems
• A Generative Model of Galactic Dust Emission Using Variational Inference
• Active Image Synthesis for Efficient Labeling
• Nonparametric Inference for Auto-Encoding Variational Bayes
• Preliminary results in using Deep Learning to emulate BLOB, a nuclear interaction model
• Greedy Hierarchical Variational Autoencoders for Large-Scale Video Prediction
• Assisted Sound Sample Generation with Musical Conditioning in Adversarial Auto-Encoders
• Multimodal Generative Models for Compositional Representation Learning
• Factor Analysis, Probabilistic Principal Component Analysis, Variational Inference, and Variational Autoencoder: Tutorial and Survey
• Performing Co-Membership Attacks Against Deep Generative Models
• Failure Modes of Variational Autoencoders and Their Effects on Downstream Tasks
• Learning Neurosymbolic Generative Models via Program Synthesis
• Diffusion-Weighted Magnetic Resonance Brain Images Generation with Generative Adversarial Networks and Variational Autoencoders: A Comparison Study
• Multi-modal data generation with a deep metric variational autoencoder
• MAD-VAE: Manifold Awareness Defense Variational Autoencoder
• Gradient Origin Networks
• Domain Fingerprints for No-reference Image Quality Assessment
• Generative Models for Security: Attacks, Defenses, and Opportunities
• LDC-VAE: A Latent Distribution Consistency Approach to Variational AutoEncoders
• Characterizing and Avoiding Problematic Global Optima of Variational Autoencoders
• Cauchy-Schwarz Regularized Autoencoder
• D2C: Diffusion-Denoising Models for Few-shot Conditional Generation
• Scalable Modeling of Spatiotemporal Data using the Variational Autoencoder: an Application in Glaucoma
• Data Augmentation for Deep Candlestick Learner
• To Regularize or Not To Regularize? The Bias Variance Trade-off in Regularized AEs
• Benefiting Deep Latent Variable Models via Learning the Prior and Removing Latent Regularization
• Bridging the ELBO and MMD
• Momentum Contrastive Autoencoder: Using Contrastive Learning for Latent Space Distribution Matching in WAE
• Decoupled Variational Embedding for Signed Directed Networks
• The Effects of Invertibility on the Representational Complexity of Encoders in Variational Autoencoders
• Learning Deep-Latent Hierarchies by Stacking Wasserstein Autoencoders
• Inverse Graphics: Unsupervised Learning of 3D Shapes from Single Images
• Noise Contrastive Variational Autoencoders
• Challenging the Semi-Supervised VAE Framework for Text Classification
• Improve variational autoEncoder with auxiliary softmax multiclassifier