Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Model Selection in Bayesian Neural Networks via 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 horseshoe prior over node pre-activations instead of Gaussian weight prior | Bayesian neural network · 1-D cubic regression (50/100/1000 units) | robustness to under-fitting | Improved |
| use non-centered parameterization for the horseshoe prior (ablation) | HS-BNN · 2-D synthetic classification, 15-unit layer | structure recovery (units pruned): prunes away all but two units | Improved |
| use horseshoe prior instead of Gaussian prior | BNN · MNIST | test error | Improved |
| use horseshoe prior instead of scale-mixture weight prior | SM-BNN · MNIST | test error | No clear effect |
| use horseshoe node prior instead of scale-mixture weight prior | SM-BNN · MNIST | node sparsity | Improved |
| use structured matrix-variate approximation (VMG) instead of factorized HS-BNN | HS-BNN · MNIST | test error | Improved |
| use horseshoe prior instead of Gaussian prior | BNN · gesture recognition (supplement) | error rate | Improved |
| use HS-BNN instead of VMG | VMG · UCI Energy | RMSE: 0.54 -> 1.99 | Hurt |
| use HS-BNN instead of VMG | VMG · UCI Yacht | RMSE: 0.70 -> 1.58 | Hurt |
| use HS-BNN instead of VMG | VMG · UCI Naval | test log-likelihood: +2.46 -> +5.52 | Improved |
| use HS-BNN instead of VMG | VMG · UCI Boston | RMSE: 2.72 -> 3.32 | Hurt |
| use HS-BNN instead of VMG | VMG · UCI Power Plant | RMSE: 4.04 -> 4.03 | No clear effect |
Closest papers by meaning
- Structured Variational Learning of Bayesian Neural Networks with Horseshoe Priors (ICML 2018 · similarity 0.95)
- A Study of Bayesian Neural Network Surrogates for Bayesian Optimization (ICLR 2024 · similarity 0.87)
- Exact marginal prior distributions of finite Bayesian neural networks (NeurIPS 2021 · similarity 0.87)
- Posterior Concentrations of Fully-Connected Bayesian Neural Networks with General Priors on the Weights (JMLR 2025 · similarity 0.87)
- Martingale Posterior Neural Networks for Fast Sequential Decision Making (NeurIPS 2025 · similarity 0.86)
- Understanding Priors in Bayesian Neural Networks at the Unit Level (ICML 2019 · similarity 0.86)
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.
