Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Fast Expectation Propagation for Heteroscedastic, Lasso-Penalized, and Quantile Regression
Jackson Zhou, John T. Ormerod, Clara Grazian
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 analytic integral reduction (univariate quadrature) instead of bivariate quadrature (ablation) | Gaussian EP, heteroscedastic regression · Energy big data | run time (s): 630.3 -> 11.1 | Improved |
| use analytic integral reduction instead of bivariate quadrature (ablation) | Gaussian EP, quantile regression · Energy big data | run time (s): 1234.5 -> 35.2 | Improved |
| use analytic integral reduction instead of bivariate quadrature (ablation) | Gaussian EP, lasso regression · Energy big data, conservative settings | run time (s): 6923.2 -> 33.3 | Improved |
| use analytic integral reduction instead of bivariate quadrature (ablation) | Gaussian EP, heteroscedastic regression · Energy big data, conservative settings | run time (s): 6994.1 -> 28.7 | Improved |
| use analytic integral reduction instead of bivariate quadrature (ablation) | Gaussian EP, quantile regression · Energy big data, conservative settings | run time (s): 6861.8 -> 78.1 | Improved |
| use analytic reduction (EP-1D) instead of bivariate quadrature (ablation) | Gaussian EP · Heteroscedastic simulations and benchmarks | L1 accuracy: 99.1 ± 0.0 vs 99.1 ± 0.0 | No clear effect |
| use EP-1D instead of Laplace approximation | heteroscedastic regression · Energy big data | beta1 L1 accuracy (%): 81.9 -> 99.1 | Improved |
| use EP-1D instead of Pathfinder | heteroscedastic regression · Energy big data | beta1 L1 accuracy (%): 25.5 -> 99.1 | Improved |
| increase Pathfinder returned samples from 100 to 1000 or 10,000 (ablation) | Pathfinder · Energy big data, heteroscedastic | beta1 L1 accuracy (%): 25.5 -> 0.7 -> 9.8 | No clear effect |
| use EP-1D instead of MFVB when n < p | lasso regression · Lasso simulation Setting 3 | beta L1 accuracy (%): 89.9 -> 91.3 | Improved |
| use EP-1D instead of MFVB | quantile regression · Quantile benchmarks IgG | beta L1 accuracy (%): 65.8 -> 99.1 | Improved |
| use EP-1D instead of MFVB | lasso regression · Energy big data | kappa L1 accuracy (%): 98.9 -> 77.7 | Hurt |
Closest papers by meaning
- Fearless Stochasticity in Expectation Propagation (NeurIPS 2024 · similarity 0.91)
- Expectation Propagation as a Way of Life: A Framework for Bayesian Inference on Partitioned Data (JMLR 2020 · similarity 0.90)
- State Space Expectation Propagation: Efficient Inference Schemes for Temporal Gaussian Processes (ICML 2020 · similarity 0.86)
- Quantile Propagation for Wasserstein-Approximate Gaussian Processes (NeurIPS 2020 · similarity 0.86)
- A Stochastic Path Integral Differential EstimatoR Expectation Maximization Algorithm (NeurIPS 2020 · similarity 0.86)
- On the Global Convergence of (Fast) Incremental Expectation Maximization Methods (NeurIPS 2019 · 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.
