Approximation Based Variance Reduction for Reparameterization Gradients
arXiv:2007.14634
Abstract
Flexible variational distributions improve variational inference but are harder to optimize. In this work we present a control variate that is applicable for any reparameterizable distribution with known mean and covariance matrix, e.g. Gaussians with any covariance structure. The control variate is based on a quadratic approximation of the model, and its parameters are set using a double-descent scheme by minimizing the gradient estimator's variance. We empirically show that this control variate leads to large improvements in gradient variance and optimization convergence for inference with non-factorized variational distributions.
arXiv commentNeural Information Processing Systems (NeurIPS 2020)
References in corpus (14)
- Adam: A Method for Stochastic Optimization
- Auto-Encoding Variational Bayes
- Categorical Reparameterization with Gumbel-Softmax
- The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables
- Neural Variational Inference and Learning in Belief Networks
- REBAR: Low-variance, unbiased gradient estimates for discrete latent variable models
- Variational inference for Monte Carlo objectives
- Variational Bayesian Inference with Stochastic Search
- Improving Variational Auto-Encoders using Householder Flow
- Monte Carlo Gradient Estimation in Machine Learning
- Reducing Reparameterization Gradient Variance
- Advances in Variational Inference
- Using Large Ensembles of Control Variates for Variational Inference
- Amortized variance reduction for doubly stochastic objectives