Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Structured Variational Learning of Bayesian Neural Networks with Horseshoe Priors
Soumya Ghosh, Jiayu Yao, Finale Doshi-Velez
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 a regularized horseshoe prior instead of a standard horseshoe prior (ablation) | Bayesian neural networks with factorized variational approximation · UCI regression datasets | predictive performance | Improved |
| Add a structured variational approximation to capture dependencies among weights and unit scales (ablation) | Regularized horseshoe BNN · UCI regression datasets | shrinkage toward zero | Improved |
| Add a structured variational approximation instead of a factorized approximation (ablation) | Regularized horseshoe BNN · UCI regression datasets | predictive performance | No clear effect |
| Use a structured approximation rather than a semi-structured approximation (ablation) | Regularized horseshoe BNN · UCI regression datasets | predictive performance | Improved |
| Use a factorized approximation rather than a tied-factorized approximation (ablation) | Regularized horseshoe BNN · UCI regression datasets | predictive performance | Improved |
| Add a pruning rule based on posterior probability that the unit scale is below a threshold | Structured regularized horseshoe BNN · UCI regression datasets | model compression | Improved |
| Fine-tune the pruned model by updating weight means while holding other parameters fixed (ablation) | Pruned structured regularized horseshoe BNN · UCI regression datasets | predictive performance | No clear effect |
| Use a regularized horseshoe BNN with structured approximation instead of a standard BNN | BNN x-500-y · 2D Map | Test RMSE: 0.187 -> 0.058 | Improved |
| Use a regularized horseshoe BNN with structured approximation instead of a standard BNN | BNN x-100-100-y · 2D Map | Test RMSE: 0.089 -> 0.061 | Improved |
| Use a regularized horseshoe BNN with structured approximation instead of a standard BNN | BNN x-500-y · 2D Map | Avg. Reward: 975.386 -> 995.416 | Improved |
| Use a regularized horseshoe BNN with structured approximation instead of a standard BNN | BNN x-100-100-y · 2D Map | Avg. Reward: 966.716 -> 992.893 | Improved |
| Use a regularized horseshoe BNN with structured approximation instead of a standard BNN | BNN x-500-y · Acrobot | Test RMSE: 0.924 -> 0.558 | Improved |
Closest papers by meaning
- Model Selection in Bayesian Neural Networks via Horseshoe Priors (JMLR 2019 · similarity 0.95)
- Deterministic Variational Inference for Robust Bayesian Neural Networks (ICLR 2019 · similarity 0.88)
- Posterior and Variational Inference for Deep Neural Networks with Heavy-Tailed Weights (JMLR 2025 · similarity 0.88)
- Bayesian Optimization via Continual Variational Last Layer Training (ICLR 2025 · similarity 0.86)
- Martingale Posterior Neural Networks for Fast Sequential Decision Making (NeurIPS 2025 · similarity 0.86)
- A Study of Bayesian Neural Network Surrogates for Bayesian Optimization (ICLR 2024 · similarity 0.85)
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.
