Implicit Reparameterization Gradients

Abstract

By providing a simple and efficient way of computing low-variance gradients of continuous random variables, the reparameterization trick has become the technique of choice for training a variety of latent variable models. However, it is not applicable to a number of important continuous distributions. We introduce an alternative approach to computing reparameterization gradients based on implicit differentiation and demonstrate its broader applicability by applying it to Gamma, Beta, Dirichlet, and von Mises distributions, which cannot be used with the classic reparameterization trick. Our experiments show that the proposed approach is faster and more accurate than the existing gradient estimators for these distributions.

arXiv commentNeurIPS 2018

Cited by in corpus (63)

• The Conditional Entropy Bottleneck
• Monte Carlo Gradient Estimation in Machine Learning
• Pathwise Derivatives Beyond the Reparameterization Trick
• GOT: An Optimal Transport framework for Graph comparison
• Continuous Hierarchical Representations with Poincaré Variational Auto-Encoders
• Auto-FedAvg: Learnable Federated Averaging for Multi-Institutional Medical Image Segmentation
• FLAMBE: Structural Complexity and Representation Learning of Low Rank MDPs
• A Tutorial on Deep Latent Variable Models of Natural Language
• Diffusion Variational Autoencoders
• Sinkhorn AutoEncoders
• Decision-Making with Auto-Encoding Variational Bayes
• not-MIWAE: Deep Generative Modelling with Missing not at Random Data
• The Power Spherical distribution
• Differentiable Unsupervised Feature Selection based on a Gated Laplacian
• Dirichlet Variational Autoencoder for Text Modeling
• A Discrete Variational Recurrent Topic Model without the Reparametrization Trick
• f-Divergence Variational Inference
• Explaining Visual Models by Causal Attribution
• Flexible Prior Elicitation via the Prior Predictive Distribution
• Hierarchical Indian Buffet Neural Networks for Bayesian Continual Learning
• Accelerating Monte Carlo Bayesian Inference via Approximating Predictive Uncertainty over Simplex
• Variational Variance: Simple, Reliable, Calibrated Heteroscedastic Noise Variance Parameterization
• BasisVAE: Translation-invariant feature-level clustering with Variational Autoencoders
• Storchastic: A Framework for General Stochastic Automatic Differentiation
• Radial and Directional Posteriors for Bayesian Neural Networks
• Variational Bayesian Decision-making for Continuous Utilities
• Factor Analysis, Probabilistic Principal Component Analysis, Variational Inference, and Variational Autoencoder: Tutorial and Survey
• Tractable structured natural gradient descent using local parameterizations
• Hyperprior Induced Unsupervised Disentanglement of Latent Representations
• Variational Auto-encoder Based Bayesian Poisson Tensor Factorization for Sparse and Imbalanced Count Data
• The Thermodynamic Variational Objective
• Handling the Positive-Definite Constraint in the Bayesian Learning Rule
• Scalable Bayesian dynamic covariance modeling with variational Wishart and inverse Wishart processes
• A unified view of likelihood ratio and reparameterization gradients and an optimal importance sampling scheme
• OCEAN: Online Task Inference for Compositional Tasks with Context Adaptation
• Self-supervised Neural Audio-Visual Sound Source Localization via Probabilistic Spatial Modeling
• Variational Marginal Particle Filters
• Automatic Differentiation Variational Inference with Mixtures
• Neuron ranking -- an informed way to condense convolutional neural networks architecture
• Instance-Dependent Partial Label Learning
• Dirichlet Pruning for Neural Network Compression
• Pathwise Derivatives for Multivariate Distributions
• Embedded-model flows: Combining the inductive biases of model-free deep learning and explicit probabilistic modeling
• A New Distribution on the Simplex with Auto-Encoding Applications
• Copula-like Variational Inference
• Generalized Transformation-based Gradient
• Amortised Learning by Wake-Sleep
• Sparse Communication via Mixed Distributions
• Don't Wait, Just Weight: Improving Unsupervised Representations by Learning Goal-Driven Instance Weights
• Practical Bayesian Learning of Neural Networks via Adaptive Optimisation Methods
• Deep Direct Likelihood Knockoffs
• Improving Relational Regularized Autoencoders with Spherical Sliced Fused Gromov Wasserstein
• Semi-supervised Stochastic Multi-Domain Learning using Variational Inference
• Continuous normalizing flows on manifolds
• Bayesian Sparsification Methods for Deep Complex-valued Networks
• Stein Latent Optimization for Generative Adversarial Networks
• Stochastic Contrastive Learning
• Pathologies in priors and inference for Bayesian transformers
• Statistical Guarantees for Transformation Based Models with Applications to Implicit Variational Inference
• Perturbative estimation of stochastic gradients
• Remaining Useful Life Estimation Under Uncertainty with Causal GraphNets
• Learning Consistent Deep Generative Models from Sparse Data via Prediction Constraints
• Meta-Learning with Variational Bayes