Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Simulation-based inference
Estimates posteriors for simulators using learned statistics and active sequential estimation.
118 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Uncertainty and Gaussian processes, keyword cluster “bayesian inference · posterior · priors”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
28 papers in 2023–24 and 33 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 10.3% to 10.7%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Coresets for Scalable Bayesian Logistic Regression (NeurIPS 2016 · 113 citations)
- 2017 · Flexible statistical inference for mechanistic models of neural dynamics (NeurIPS 2017 · 139 citations)
- 2018 · Fast Information-theoretic Bayesian Optimisation (ICML 2018 · 26 citations)
- 2019 · MFBO-SSM: Multi-Fidelity Bayesian Optimization for Fast Inference in State-Space Models (AAAI 2019 · 67 citations)
- 2020 · Variational Bayesian Monte Carlo with Noisy Likelihoods (NeurIPS 2020 · 31 citations)
- 2021 · Learning an Explicit Weighting Scheme for Adapting Complex HSI Noise (CVPR 2021 · 14 citations)
- 2022 · Truncated proposals for scalable and hassle-free simulation-based inference (NeurIPS 2022 · 25 citations)
- 2023 · Learning Robust Statistics for Simulation-based Inference under Model Misspecification (NeurIPS 2023 · 9 citations)
- 2024 · All-in-one simulation-based inference (ICML 2024 · 5 citations)
- 2025 · Addressing Misspecification in Simulation-based Inference through Data-driven Calibration (ICML 2025 · 3 citations)
- 2026 · Compositional amortized inference for large-scale hierarchical Bayesian models (ICLR 2026 · 1 citation)
Earliest papers here
- Coresets for Scalable Bayesian Logistic Regression (NeurIPS 2016 · 113 citations)
- Estimating Cosmological Parameters from the Dark Matter Distribution (ICML 2016 · 17 citations)
- DR-ABC: Approximate Bayesian Computation with Kernel-Based Distribution Regression (ICML 2016 · 16 citations)
What has been tried
94 papers here report 815 measured changes (a component added, removed or swapped, compared with a baseline): helped 65%, no clear effect 14%, hurt 16%, mixed 5%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Train only on the posterior mask instead of learning all conditional distributions | Benchmark tasks · C2ST accuracy | No clear effect | All-in-one simulation-based inference (ICML 2024 · 5 citations) | |
| Add the OT step to tuning-only | Tasks A, C, D, E, F · LPP and ACAUC | Helped | Addressing Misspecification in Simulation-based Inference through Data-driven Calibration (ICML 2025 · 3 citations) | |
| Replace the MoGr model posterior with a flat categorical distribution | Additive model · data efficiency | Hurt | Simultaneous identification of models and parameters of scientific simulators (ICML 2024 · 3 citations) | |
| Use active acquisition of simulation parameters instead of non-active SNPE-C | Bernoulli GLM · C2ST | 0.749 ± 0.017 -> 0.725 ± 0.012 | Helped | Active Sequential Posterior Estimation for Sample-Efficient Simulation-Based Inference (NeurIPS 2024 · 3 citations) |
| add multi-label classification without an abundance mask | FENIR · aRMSE | 0.0875 -> 0.0967 | Hurt | Solving Spectrum Unmixing as a Multi-Task Bayesian Inverse Problem with Latent Factors for Endmember Variability (AAAI 2024 · 2 citations) |
| Add gradient rescaling and gradient projection | Neural SBI training · Optimization stability | Helped | Multilevel neural simulation-based inference (NeurIPS 2025 · 1 citation) | |
| Use KL minimization instead of TB | Set generation, α=1.00 · TV distance | 0.21 ± 0.06 -> 0.13 ± 0.03 | Helped | Streaming Bayes GFlowNets (NeurIPS 2024 · 1 citation) |
| add the divergence loss for particle interaction (α=0.02 rather than α=0) | DTD · Accuracy | 67.24 -> 71.70 | Helped | Promoting Ensemble Diversity with Interactive Bayesian Distributional Robustness for Fine-tuning Foundation Models (ICML 2025 · 0 citations) |
| Filter simulations by relevance when using 10^5 simulations | SBI benchmark tasks · C2ST accuracy | Helped | Effortless, Simulation-Efficient Bayesian Inference using Tabular Foundation Models (NeurIPS 2025 · 0 citations) | |
| Add sufficiently many Fourier modes to the FNO blocks | Linear Gaussian · SWD | Helped | FNOPE: Simulation-based inference on function spaces with Fourier Neural Operators (NeurIPS 2025 · 0 citations) | |
| Jointly train imputation and inference rather than train them separately | GLM, GLU at 10%, 25%, and 60% missing · RMSE | 0.67 -> 0.65; 0.99 -> 0.93; 1.31 -> 1.27 (GLM); 0.43 -> 0.41; 0.45 -> 0.43; 0.58 -> 0.56 (GLU) | Helped | Robust Simulation-Based Inference under Missing Data via Neural Processes (ICLR 2025 · 0 citations) |
| Increase the number of quantile bins | Two Moons (TM), large simulation budgets · C2ST | Helped | Simulation-Based Inference with Quantile Regression (ICML 2024 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 18 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Baselines
- NPE · 3 of 18 papers
- NRE · 2 of 18 papers
- Random design · 2 of 18 papers
- VPCE · 2 of 18 papers
- RL-BOED · 2 of 18 papers
- Simformer · 2 of 18 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 61 papers; from OpenAlex and the papers' affiliations).
- University College London · 6 papers
- University of Tuebingen · 4 papers
- University of Oxford · 3 papers
- Princeton University · 3 papers
- University of British Columbia · 3 papers
- Duke University · 3 papers
- Aalto University · 3 papers
- Eberhard-Karls-Universität Tübingen · 3 papers
- Imperial College London · 2 papers
- University Tübingen · 2 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Nonparametric and PAC-Bayes Bayesian learning (130 papers)
- Variational Bayesian deep learning (105 papers)
- Approximate inference in Bayesian neural networks (93 papers)
- Bayesian view of meta-learning (63 papers)
Elsewhere in the topic
- Priors and functional variational uncertainty (112 papers)
- Sparse variational Gaussian processes (92 papers)
- PAC-Bayes and decoupled GP learning (83 papers)
- Epistemic versus aleatoric decomposition (75 papers)
- Single-model predictive uncertainty (74 papers)
- Bayesian decision-making under uncertainty (57 papers)
- Infinite-width networks as Gaussian processes (49 papers)
- Regression uncertainty quantification (43 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).