Supercharging Simulation-Based Inference for Bayesian Optimal Experimental Design
Not on the map. Why?
Abstract (arXiv)
Bayesian optimal experimental design (BOED) seeks to maximize the expected information gain (EIG) of experiments. This requires a likelihood estimate, which in many settings is intractable. Simulation-based inference (SBI) provides powerful tools for this regime. However, existing work explicitly connecting SBI and BOED is restricted to a single contrastive EIG bound. We show that the EIG admits multiple formulations which can directly leverage modern SBI density estimators, encompassing neural posterior, likelihood, and ratio estimation. Building on this perspective, we define a novel EIG estimator using neural likelihood estimation. Further, we identify optimization as a key bottleneck of gradient based EIG maximization and show that a simple multi-start parallel gradient ascent procedure can substantially improve reliability and performance. With these innovations, our SBI-based BOED methods are able to match or outperform by up to 22% existing state-of-the-art approaches across standard BOED benchmarks.
Key facts. Cited by 1 paper in the map (data through 13 Oct 2026). Builds on 31 of its 54 references that are in the map. First posted on arXiv 6 Feb 2026.
Builds on
31 of 54 references.
References · 31
- 2018Representation Learning with Contrastive Predictive Codingcited by 7,712
- 2015Variational Inference with Normalizing Flowscited by 2,163
- 2015Automatic differentiation in machine learning: a surveycited by 852
- 2018MINE: Mutual Information Neural Estimationcited by 682
- 2019Neural Spline Flowscited by 590
- 2019The frontier of simulation-based inferencecited by 492
- 2019Monte Carlo Gradient Estimation in Machine Learningcited by 283
- 2023Neural General Circulation Models for Weather and Climatecited by 238
- 2018Sequential Neural Likelihood: Fast Likelihood-free Inference with Autoregressive Flowscited by 217
- 2019Automatic Posterior Transformation for Likelihood-Free Inferencecited by 216
- 2017Flexible statistical inference for mechanistic models of neural dynamicscited by 162
- 2021Benchmarking Simulation-Based Inferencecited by 160
19 more
- 2023Modern Bayesian Experimental Designcited by 123
- 2019Variational Bayesian Optimal Experimental Designcited by 116
- 2021Deep Adaptive Design: Amortizing Sequential Bayesian Experimental Designcited by 105
- 2020On Contrastive Learning for Likelihood-free Inferencecited by 96
- 2017On Nesting Monte Carlo Estimatorscited by 82
- 2016Likelihood-free inference by ratio estimationcited by 80
- 2021Real-time gravitational-wave science with neural posterior estimationcited by 76
- 2021Implicit Deep Adaptive Design: Policy-Based Experimental Design without Likelihoodscited by 61
- 2019A Unified Stochastic Gradient Approach to Designing Bayesian-Optimal Experimentscited by 58
- 2020Bayesian Experimental Design for Implicit Models by Mutual Information Neural Estimationcited by 56
- 2022Optimizing Sequential Experimental Design with Deep Reinforcement Learningcited by 46
- 2022Truncated proposals for scalable and hassle-free simulation-based inferencecited by 40
- 2018Efficient Bayesian Experimental Design for Implicit Modelscited by 34
- 2021Sequential Neural Posterior and Likelihood Approximationcited by 30
- 2024sbi reloaded: a toolkit for simulation-based inference workflowscited by 23
- 2022neos: End-to-End-Optimised Summary Statistics for High Energy Physicscited by 10
- 2022Differentiable Matrix Elements with MadJaxcited by 8
- 2024Active Sequential Posterior Estimation for Sample-Efficient Simulation-Based Inferencecited by 5
- 2025Optimizing Likelihoods via Mutual Information: Bridging Simulation-Based Inference and Bayesian Optimal Experimental Designcited by 2
Built on by
Citing papers per year: 2026: 1.
1 paper.