Neural Variational Inference and Learning in Belief Networks
arXiv:1402.0030
Abstract
Highly expressive directed latent variable models, such as sigmoid belief networks, are difficult to train on large datasets because exact inference in them is intractable and none of the approximate inference methods that have been applied to them scale well. We propose a fast non-iterative approximate inference method that uses a feedforward network to implement efficient exact sampling from the variational posterior. The model and this inference network are trained jointly by maximizing a variational lower bound on the log-likelihood. Although the naive estimator of the inference model gradient is too high-variance to be useful, we make it practical by applying several straightforward model-independent variance reduction techniques. Applying our approach to training sigmoid belief networks and deep autoregressive networks, we show that it outperforms the wake-sleep algorithm on MNIST and achieves state-of-the-art results on the Reuters RCV1 document dataset.
References in corpus (5)
Cited by in corpus (91)
- An Introduction to Variational Autoencoders
- Variational Inference with Normalizing Flows
- Scalable Bayesian Optimization Using Deep Neural Networks
- Generative Moment Matching Networks
- Variational Autoencoder for Deep Learning of Images, Labels and Captions
- Markov Chain Monte Carlo and Variational Inference: Bridging the Gap
- Opportunities and challenges for quantum-assisted machine learning in near-term quantum computers
- Learning Semantic Segmentation of Large-Scale Point Clouds with Random Sampling
- Deep Learning: A Bayesian Perspective
- On the Convergence of Stochastic Gradient MCMC Algorithms with High-Order Integrators
- BIVA: A Very Deep Hierarchy of Latent Variables for Generative Modeling
- Q-Prop: Sample-Efficient Policy Gradient with An Off-Policy Critic
- One-Shot Generalization in Deep Generative Models
- WHAI: Weibull Hybrid Autoencoding Inference for Deep Topic Modeling
- Deep Temporal Sigmoid Belief Networks for Sequence Modeling
- Classification Aware Neural Topic Model and its Application on a New COVID-19 Disinformation Corpus
- -means as a variational EM approximation of Gaussian mixture models
- Data Generation as Sequential Decision Making
- Deep Conversational Recommender in Travel
- Exponentially Increasing the Capacity-to-Computation Ratio for Conditional Computation in Deep Learning
- Hierarchical Quantized Autoencoders
- Liberty or Depth: Deep Bayesian Neural Nets Do Not Need Complex Weight Posterior Approximations
- EM-like Learning Chaotic Dynamics from Noisy and Partial Observations
- Max-margin Deep Generative Models
- Gradient Estimation with Stochastic Softmax Tricks
- Learning Wake-Sleep Recurrent Attention Models
- Mean-Field Networks
- Deformable Generator Networks: Unsupervised Disentanglement of Appearance and Geometry
- Fast Second-Order Stochastic Backpropagation for Variational Inference
- A RAD approach to deep mixture models
- Hierarchical Variational Models
- A Review of Learning with Deep Generative Models from Perspective of Graphical Modeling
- Estimating Gradients for Discrete Random Variables by Sampling without Replacement
- Reparameterization Gradients through Acceptance-Rejection Sampling Algorithms
- Bidirectional Helmholtz Machines
- GP-select: Accelerating EM using adaptive subspace preselection
- A Discrete Variational Recurrent Topic Model without the Reparametrization Trick
- Cascaded Text Generation with Markov Transformers
- Learning document embeddings along with their uncertainties
- Learning to Predict Explainable Plots for Neural Story Generation
- Efficient Marginalization of Discrete and Structured Latent Variables via Sparsity
- Incremental Text to Speech for Neural Sequence-to-Sequence Models using Reinforcement Learning
- DisARM: An Antithetic Gradient Estimator for Binary Latent Variables
- Learning Discrete Structured Representations by Adversarially Maximizing Mutual Information
- Storchastic: A Framework for General Stochastic Automatic Differentiation
- Learning Deep Generative Models with Doubly Stochastic MCMC
- Clinical Tagging with Joint Probabilistic Models
- Automatic Relevance Determination For Deep Generative Models
- Consistent Sparse Deep Learning: Theory and Computation
- Amortized Bethe Free Energy Minimization for Learning MRFs
- Local Expectation Gradients for Doubly Stochastic Variational Inference
- Variational Generative Stochastic Networks with Collaborative Shaping
- Unsupervised Recurrent Neural Network Grammars
- Learning to learn generative programs with Memoised Wake-Sleep
- The Thermodynamic Variational Objective
- Iterative Amortized Policy Optimization
- GO Gradient for Expectation-Based Objectives
- Select and Attend: Towards Controllable Content Selection in Text Generation
- Task-adaptive Neural Process for User Cold-Start Recommendation
- VQ-DRAW: A Sequential Discrete VAE
- Divergence Triangle for Joint Training of Generator Model, Energy-based Model, and Inference Model
- Path Sample-Analytic Gradient Estimators for Stochastic Binary Networks
- Rao-Blackwellizing the Straight-Through Gumbel-Softmax Gradient Estimator
- Cross-Modal Generative Augmentation for Visual Question Answering
- Modeling Item Response Theory with Stochastic Variational Inference
- GSNs : Generative Stochastic Networks
- Monte Carlo Filtering Objectives: A New Family of Variational Objectives to Learn Generative Model and Neural Adaptive Proposal for Time Series
- Adaptive Correlated Monte Carlo for Contextual Categorical Sequence Generation
- Gaussian Process Bandit Optimization of the Thermodynamic Variational Objective
- Deep Autoencoding Topic Model with Scalable Hybrid Bayesian Inference
- An Introduction to Spiking Neural Networks: Probabilistic Models, Learning Rules, and Applications
- Wasserstein Learning of Determinantal Point Processes
- CARMS: Categorical-Antithetic-REINFORCE Multi-Sample Gradient Estimator
- Reconciling the Discrete-Continuous Divide: Towards a Mathematical Theory of Sparse Communication
- Learning Energy-Based Models as Generative ConvNets via Multi-grid Modeling and Sampling
- Read, Attend and Comment: A Deep Architecture for Automatic News Comment Generation
- Generalized Transformation-based Gradient
- A Rule for Gradient Estimator Selection, with an Application to Variational Inference
- Reliable Categorical Variational Inference with Mixture of Discrete Normalizing Flows
- Variational Sequential Labelers for Semi-Supervised Learning
- Sparse Communication via Mixed Distributions
- A Tale of Three Probabilistic Families: Discriminative, Descriptive and Generative Models
- Hybrid Memoised Wake-Sleep: Approximate Inference at the Discrete-Continuous Interface
- Learning to Organize Knowledge and Answer Questions with N-Gram Machines
- Curious Explorer: a provable exploration strategy in Policy Learning
- MirrorNet: A Deep Bayesian Approach to Reflective 2D Pose Estimation from Human Images
- Meta-Learning an Inference Algorithm for Probabilistic Programs
- Learning Opinion Summarizers by Selecting Informative Reviews
- Which is Making the Contribution: Modulating Unimodal and Cross-modal Dynamics for Multimodal Sentiment Analysis
- Approximation Based Variance Reduction for Reparameterization Gradients
- Representation Learning: A Statistical Perspective