ENFR: Evidence Networks with Flow-Matched References for Amortized Simulation-Based Model Inference
The evidence, $\mathcal{Z} = p(x | \mathcal{M})$, of a data-generating model $\mathcal{M}$ is a powerful tool in Bayesian frameworks: it enables model inference, model misspecification tests, and goodness-of-fit assessment as an alternative to standard discrepancy statistics, e.g. $χ^2$. While versatile, the evidence is notoriously difficult to compute, even when the likelihood
Reading in Plain English mode — toggle below to switch
The evidence, $\mathcal{Z} = p(x | \mathcal{M})$, of a data-generating model $\mathcal{M}$ is a powerful tool in Bayesian frameworks: it enables model inference, model misspecification tests, and goodness-of-fit assessment as an alternative to standard discrepancy statistics, e.g. $χ^2$. While versatile, the evidence is notoriously difficult to compute, even when the likelihood
This paper introduces a novel approach for estimating model evidence, a crucial component for robust Bayesian inference, model validation, and goodness-of-fit assessment. For ML engineers, understanding how to efficiently compute evidence can lead to more trustworthy and interpretable models, especially in simulation-based inference where traditional likelihoods are intractable.