A1807
Title: Simulation-based inference via structured score matching
Authors: Haoyu Jiang - UIUC (United States)
Yuexi Wang - UIUC (United States)
Yun Yang - University of Maryland, College Park (United States) [presenting]
Abstract: Simulation-based inference (SBI) provides an effective framework for statistical analysis when the likelihood is intractable but model simulations are available. A unified SBI framework is presented that leverages structured score matching for both frequentist and Bayesian inference. On the frequentist side, a likelihood-free approach combines score matching with gradient-based optimization and bootstrap procedures for parameter estimation and uncertainty quantification. On the Bayesian side, score matching is integrated with Langevin dynamics to efficiently explore complex posterior landscapes in moderate to high-dimensional settings. In both cases, tailored score-matching estimators and architectural regularizations are designed that embed the statistical structure of log-likelihood scores, which improves estimation accuracy and scalability.
