Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
For high-dimensional hierarchical models, consider exchangeability of effects across covariates instead of across datasets
Brian L. Trippe, Hilary Finucane, Tamara Broderick
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 |
|---|---|---|---|
| model effects as exchangeable across covariates (ECov) instead of across datasets (EData) | hierarchical Bayesian regression · simulation, Q=10, D > Q | parameter estimation error | Improved |
| model effects as exchangeable across datasets (EData) when datasets outnumber covariates | hierarchical Bayesian regression · simulation, Q=10, D < Q | parameter estimation error | Improved |
| estimate Sigma by maximum marginal likelihood (PSD-constrained) instead of moment matching | ECov and EData · simulation | parameter estimation error | Improved |
| share strength across datasets via ECov rather than fitting each dataset independently (ID) | ECov · simulation with independent effects | parameter estimation error | No clear effect |
| ECov vs least squares and EData when effects are independent across datasets | ECov · simulation, high D | parameter estimation error | Improved |
| use ECov instead of EData, independent Bayes and least squares | ECov · law enforcement, blog posts, CIFAR10 | held-out prediction error | Improved |
| model dataset heterogeneity instead of pooling all datasets | ECov · three real applications | held-out prediction error | Improved |
| ECov moment-matching estimator replaces least squares when D > 2Q+2 | least squares · theory, orthogonal design | squared error risk | Improved |
| EData moment-matching estimator versus least squares when D is large relative to Q | least squares · theory, orthogonal design | squared error risk | Hurt |
| replace moment-matching with maximum marginal likelihood (positive-part shrinkage) for ECov | ECov · theory, D > Q+1 | squared error risk | Improved |
| use a Sigma estimate with factor 1/(D-Q-1) instead of the maximum marginal likelihood one | ECov positive-part estimator · unreported experiments | estimation error | Hurt |
Closest papers by meaning
- Bivariate Causal Discovery using Bayesian Model Selection (ICML 2024 · similarity 0.84)
- Continuous Bayesian Model Selection for Multivariate Causal Discovery (ICML 2025 · similarity 0.83)
- Hierarchical and Density-based Causal Clustering (NeurIPS 2024 · similarity 0.83)
- Doubly robust identification of treatment effects from multiple environments (ICLR 2025 · similarity 0.83)
- Differentiable Causal Discovery for Latent Hierarchical Causal Models (ICLR 2025 · similarity 0.83)
- Do Finetti: On Causal Effects for Exchangeable Data (NeurIPS 2024 · similarity 0.83)
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.
