Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Bayesian Leave-One-Out Cross-Validation Approximations for Gaussian Latent Variable Models
Aki Vehtari, Tommi Mononen, Ville Tolvanen, Tuomas Sivula, Ole Winther
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use LA-LOO (Laplace tilted-marginal quadrature) instead of Gaussian-marginal quadrature | LOO approximation · Australian/Ionosphere/Sonar GP classification | bias vs exact LOO: Q-LOO-LA-G fails (NA/-6732 on Sonar) vs LA-LOO -0.2 | Improved |
| use LA-LOO instead of WAIC_V-LA-L | LOO approximation · Sonar | bias: 17 -> -0.2 | Improved |
| use EP-LOO instead of WAIC_V-EP-L | LOO approximation · Australian | bias: 54 -> 1.6 | Improved |
| use WAIC_V (third-order) instead of WAIC_G | WAIC approximation · Sonar, LA-L marginals | bias: 26 -> 17 | Improved |
| replace Gaussian marginal with LA-L tilted marginal in WAIC | WAIC_V · Ionosphere | bias: -616 -> 16 | Improved |
| add global correction LA-CM2 to the marginal approximation | LOO approximation · Ionosphere | bias vs exact LOO: LA-LOO -23 vs Q-LOO-LA-CM2 9.0 | Mixed |
| add EP-FACT global correction to EP-LOO (ablation) | EP-LOO · Sonar | bias: -0.5 -> -0.64 | No clear effect |
| quadrature with EP-FACT marginals instead of EP-LOO | Q-LOO-EP-FACT · Leukemia | bias: -0.0 -> 4.3 | Hurt |
| increase model flexibility (smaller length scale) for approximations other than LA-LOO/EP-LOO (ablation) | LOO approximations · GP data sets | accuracy: fail when p_eff/n larger than 2%-5% | Hurt |
| add importance weighting to CCD integration of hyperparameters (ablation) | LA-LOO · Ripley | bias: 0.8 -> 0.18 | Improved |
| add importance weighting to CCD integration of hyperparameters (ablation) | LA-LOO · Leukemia | bias: 4.8 -> 0.56 | Improved |
| add importance weighting to CCD (ablation) | LA-LOO · Boston | bias: 17 -> -5.2 | Mixed |
Closest papers by meaning
- Bayesian leave-one-out cross-validation for large data (ICML 2019 · similarity 0.89)
- Hamiltonian Monte Carlo using an adjoint-differentiated Laplace approximation: Bayesian inference for latent Gaussian models and beyond (NeurIPS 2020 · similarity 0.85)
- Bayes beats Cross Validation: Efficient and Accurate Ridge Regression via Expectation Maximization (NeurIPS 2023 · similarity 0.85)
- Choice of V for V-Fold Cross-Validation in Least-Squares Density Estimation (JMLR 2016 · similarity 0.85)
- Distributed Bayesian Varying Coefficient Modeling Using a Gaussian Process Prior (JMLR 2022 · similarity 0.85)
- Learning Latent Variable Models via Jarzynski-adjusted Langevin Algorithm (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
