Truncated Marginal Neural Ratio Estimation
This work addresses the problem of trust and efficiency in simulation-based inference for scientists using complex simulators, representing an incremental improvement with novel targeting and testing features.
The paper tackles the challenge of Bayesian parameter inference in high-dimensional, intractable-likelihood simulators by proposing a neural simulation-based inference algorithm that estimates low-dimensional marginal posteriors with simulation efficiency and enables empirical posterior testing. The method demonstrates efficiency on benchmark tasks and complex posteriors, though specific numerical gains are not quantified.
Parametric stochastic simulators are ubiquitous in science, often featuring high-dimensional input parameters and/or an intractable likelihood. Performing Bayesian parameter inference in this context can be challenging. We present a neural simulation-based inference algorithm which simultaneously offers simulation efficiency and fast empirical posterior testability, which is unique among modern algorithms. Our approach is simulation efficient by simultaneously estimating low-dimensional marginal posteriors instead of the joint posterior and by proposing simulations targeted to an observation of interest via a prior suitably truncated by an indicator function. Furthermore, by estimating a locally amortized posterior our algorithm enables efficient empirical tests of the robustness of the inference results. Since scientists cannot access the ground truth, these tests are necessary for trusting inference in real-world applications. We perform experiments on a marginalized version of the simulation-based inference benchmark and two complex and narrow posteriors, highlighting the simulator efficiency of our algorithm as well as the quality of the estimated marginal posteriors.