Perturbative adaptive importance sampling for Bayesian LOO cross-validation
📰 ArXiv cs.AI
Perturbative adaptive importance sampling improves Bayesian LOO cross-validation efficiency
Action Steps
- Apply importance sampling to invert Bayesian updates for a single observation
- Use adaptive transformation to reduce variance in importance weights
- Remove a single observation to perturb the posterior and estimate the effect
- Implement perturbative adaptive importance sampling in Bayesian LOO cross-validation
Who Needs to Know This
Data scientists and machine learning researchers benefit from this approach as it enhances the efficiency of Bayesian model evaluation, while software engineers can implement the method in their models
Key Insight
💡 Adaptive transformation reduces variance in importance weights, enhancing efficiency
Share This
💡 Improve Bayesian LOO CV efficiency with perturbative adaptive importance sampling
Key Takeaways
Perturbative adaptive importance sampling improves Bayesian LOO cross-validation efficiency
Full Article
Title: Perturbative adaptive importance sampling for Bayesian LOO cross-validation
Abstract:
arXiv:2402.08151v4 Announce Type: replace-cross Abstract: Importance sampling (IS) is an efficient stand-in for model refitting in performing (LOO) cross-validation (CV) on a Bayesian model. IS inverts the Bayesian update for a single observation by reweighting posterior samples. The so-called importance weights have high variance -- we resolve this issue through adaptation by transformation. We observe that removing a single observation perturbs the posterior by $\mathcal{O}(1/n)$, motivating b
Abstract:
arXiv:2402.08151v4 Announce Type: replace-cross Abstract: Importance sampling (IS) is an efficient stand-in for model refitting in performing (LOO) cross-validation (CV) on a Bayesian model. IS inverts the Bayesian update for a single observation by reweighting posterior samples. The so-called importance weights have high variance -- we resolve this issue through adaptation by transformation. We observe that removing a single observation perturbs the posterior by $\mathcal{O}(1/n)$, motivating b
DeepCamp AI