Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Scalable Approximate MCMC Algorithms for the Horseshoe Prior
James Johndrow, Paulo Orenstein, Anirban Bhattacharya
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 |
|---|---|---|---|
| replace the old exact algorithm with the new exact algorithm | horseshoe Gibbs sampler · Simulated regression (N 200-1000, p 1000-5000) | mean squared error of beta: old has about double the MSE at the median | Improved |
| replace the old exact algorithm with the new exact algorithm | horseshoe Gibbs sampler · Simulated regression (N 200-1000, p 1000-5000) | coverage of 95 percent credible intervals: about 90 percent -> about 93 percent | Improved |
| use approximate algorithm instead of new exact algorithm | horseshoe sampler · Simulated regression (N 200-1000, p 1000-5000) | MSE and coverage | No clear effect |
| use approximate algorithm instead of new exact algorithm | horseshoe sampler · Simulated regression (log time scaling in N) | coefficient of log(N) in log(t): 1.6847 -> 0.5449 | Improved |
| use approximate algorithm instead of new exact algorithm | horseshoe sampler · Simulated regression (log time scaling in p) | coefficient of log(p) in log(t): 0.6065 -> 0.3392 | Improved |
| reduce threshold delta from 10^-2 to 10^-4 (ablation) | approximate horseshoe sampler · Simulated regression (N=1000, p=10000) | correlation of pathwise variances with exact: 0.39 -> 0.99 | Improved |
| reduce threshold delta from 10^-4 to 10^-5 (ablation) | approximate horseshoe sampler · Simulated regression (N=1000, p=10000) | correlation of pathwise variances with exact: 0.99 -> 1.00 | No clear effect |
| increase N | approximate horseshoe sampler · Simulated independent design | -log(ne) regression coefficient: -0.17 | Improved |
| correlated AR(1) design phi=0.9 instead of independent design | approximate horseshoe sampler · Simulated regression effective sample size | ne: -0.15 vs -0.17 | No clear effect |
Closest papers by meaning
- Sketching in High-Dimensional Regression With Big Data Using Gaussian Scale Mixture Priors (JMLR 2025 · similarity 0.86)
- T-LoHo: A Bayesian Regularization Model for Structured Sparsity and Smoothness on Graphs (NeurIPS 2021 · similarity 0.85)
- Scalable Bayesian Learning with posteriors (ICLR 2025 · similarity 0.85)
- Distributed Bayesian Varying Coefficient Modeling Using a Gaussian Process Prior (JMLR 2022 · similarity 0.85)
- Quasi-Monte Carlo Features for Kernel Approximation (ICML 2024 · similarity 0.84)
- Parallelizing MCMC Across the Sequence Length (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.
