Yes, but Did It Work?: Evaluating Variational Inference
arXiv:1802.02538
Abstract
While it's always possible to compute a variational approximation to a posterior distribution, it can be difficult to discover problems with this approximation. We propose two diagnostic algorithms to alleviate this problem. The Pareto-smoothed importance sampling (PSIS) diagnostic gives a goodness of fit measurement for joint distributions, while simultaneously improving the error in the estimate. The variational simulation-based calibration (VSBC) assesses the average performance of point estimates.
Appearing at International Conference on Machine Learning 2018
Cited by in corpus (21)
- Visualization in Bayesian workflow
- Stacking for Non-mixing Bayesian Computations: The Curse and Blessing of Multimodal Posteriors
- Variational Bayesian Monte Carlo with Noisy Likelihoods
- Stochastic Gradient MCMC with Repulsive Forces
- Decision-Making with Auto-Encoding Variational Bayes
- Pathfinder: Parallel quasi-Newton variational inference
- Robust, Accurate Stochastic Optimization for Variational Inference
- Fast and Accurate Estimation of Non-Nested Binomial Hierarchical Models Using Variational Inference
- Stein Neural Sampler
- Posterior inference unchained with EL_2O
- Validated Variational Inference via Practical Posterior Error Bounds
- A Parsimonious Tour of Bayesian Model Uncertainty
- Variational Bayesian Decision-making for Continuous Utilities
- Calibration procedures for approximate Bayesian credible sets
- Provable Smoothness Guarantees for Black-Box Variational Inference
- An Easy to Interpret Diagnostic for Approximate Inference: Symmetric Divergence Over Simulations
- Ensemble Model Patching: A Parameter-Efficient Variational Bayesian Neural Network
- q-Paths: Generalizing the Geometric Annealing Path using Power Means
- Validating Gaussian Process Models with Simulation-Based Calibration
- Contributions to Large Scale Bayesian Inference and Adversarial Machine Learning
- Duality between Approximate Bayesian Methods and Prior Robustness