Conservative neural posterior estimation via distributionally robust training
Simulation-based inference with neural posterior estimation (NPE) often yields overconfident and unreliable posteriors under limited simulation budgets. To address this, we propose DRO-NPE, a distributionally robust approach that replaces the standard NPE objective with a worst-case loss over a Wasserstein ambiguity set. We introduce KL-based metrics for miscoverage and miscalibration, and use these to show that the DRO-NPE objective controls overfitting and reduces posterior overconfidence. Our method is tractable, parallelisable, and readily integrates with standard normalising flows. Across benchmark SBI tasks, DRO-NPE consistently improves coverage and calibration, while narrowing the gap between empirical and population NPE loss, leading to more reliable inference in low-simulation regimes.
Code (0)
등록된 구현이 없습니다.
Similar Papers 제목 키워드 기반
Balancing Simulation-based Inference for Conservative Posteriors
Conservative inference is a major concern in simulation-based inference. It has been shown that commonly used algorithms can produce overconfident posterior approximations. Balancing has empirically proven to be an effec…
Distributionally Robust Learning for Unsupervised Domain Adaptation
We propose a distributionally robust learning (DRL) method for unsupervised domain adaptation (UDA) that scales to modern computer-vision benchmarks. DRL can be naturally formulated as a competitive two-player game be…
Density Ratio EstimationDomain AdaptationUnsupervised Domain AdaptationTowards Reliable Simulation-Based Inference with Balanced Neural Ratio Estimation
Modern approaches for simulation-based inference rely upon deep learning surrogates to enable approximate inference with computer simulators. In practice, the estimated posteriors' computational faithfulness is, however,…
Dirichlet Logistic Gaussian Processes for Evaluation of Black-Box Stochastic Systems under Complex Requirements
The requirement-driven performance evaluation of a black-box cyber-physical system (CPS) that utilizes machine learning methods has proven to be an effective way to assess the quality of the CPS. However, the distributio…
Gaussian ProcessesDROMO: Distributionally Robust Offline Model-based Policy Optimization
We consider the problem of offline reinforcement learning with model-based control, whose goal is to learn a dynamics model from the experience replay and obtain a pessimism-oriented agent under the learned model. Curren…
modelUncertainty Quantification