Research map / Probabilistic and scientific modeling / Variational and generative sampling
Neural Bayes estimators for censored inference with peaks-over-threshold models
Jordan Richards, Matthew Sainsbury-Dale, Andrew Zammit-Mangion, Raphaël Huser
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 NBE instead of censored pairwise likelihood (hmax=3) | GP spatial model · GP, tau=0.9, lambda | absolute-error test risk: 4.4 -> 3.1 | Improved |
| use NBE instead of censored pairwise likelihood (hmax=3) | MSP · MSP, tau=0.9, lambda | absolute-error test risk: 3.5 -> 2.4 | Improved |
| use NBE instead of censored pairwise likelihood (hmax=3) | IMSP · IMSP, tau=0.9, lambda | absolute-error test risk: 4.6 -> 2.6 | Improved |
| restrict pairwise likelihood to pairs within hmax=3 instead of all pairs (ablation) | censored pairwise likelihood · GP, tau=0.9, kappa | absolute-error test risk: 16.0 -> 5.7 | Improved |
| use NBE instead of quasi-Monte Carlo censored likelihood | r-Pareto process · r-Pareto, lambda | absolute-error test risk: 4.92 -> 2.29 | Improved |
| use NBE instead of quasi-Monte Carlo censored likelihood | r-Pareto process · r-Pareto, kappa | absolute-error test risk: 4.31 -> 1.87 | Improved |
| use NBE (6x6 grid) instead of quasi-Monte Carlo censored likelihood | HW process · HW, delta | absolute-error test risk: 4.80 -> 3.24 | Improved |
| use NBE (6x6 grid) instead of quasi-Monte Carlo censored likelihood | HW process · HW, lambda | absolute-error test risk: 5.82 -> 6.30 | No clear effect |
| observe HW process on 16x16 grid instead of 6x6 grid | NBE · HW, kappa | absolute-error test risk: 9.49 -> 7.42 | Improved |
| train tau-random NBE (tau sampled from Unif(0.85,0.95)) instead of tau-fixed (ablation) | NBE · HW, delta | absolute-error test risk: 3.25 -> 2.98 | Improved |
| train tau-random NBE instead of tau-fixed (ablation) | NBE · IMSP, kappa | absolute-error test risk: 2.11 -> 2.41 | Hurt |
| train tau-random NBE instead of tau-fixed (ablation) | NBE · IMSP, lambda | absolute-error test risk: 3.03 -> 2.97 | No clear effect |
Closest papers by meaning
- Mitigating the Effects of Non-Identifiability on Inference for Bayesian Neural Networks with Latent Variables (ICML 2023 · similarity 0.83)
- Conditional Distribution Function Estimation Using Neural Networks for Censored and Uncensored Data (NeurIPS 2023 · similarity 0.83)
- Amortized Simulation-Based Inference in Generalized Bayes via Neural Posterior Estimation (ICML 2026 · similarity 0.82)
- Leveraging Self-Consistency for Data-Efficient Amortized Bayesian Inference (ICML 2024 · similarity 0.81)
- Neural Conditional Probability for Uncertainty Quantification (NeurIPS 2024 · similarity 0.80)
- A Study of Bayesian Neural Network Surrogates for Bayesian Optimization (ICLR 2024 · similarity 0.80)
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.
