Enables Bayesian model selection and joint posterior inference over combinatorial spaces of up to billions of simulator model instantiations.
arXiv · March 17, 2026 · 2603.15292
The Takeaway
Traditional Bayesian selection is intractable for large families of simulators; PRISM uses an encoder-decoder framework with test-time complexity control to navigate these spaces. This is a significant advance for scientific ML, allowing researchers to automate the selection of the most parsimonious physical model from billions of possibilities.
From the abstract
Simulation plays a central role in scientific discovery. In many applications, the bottleneck is no longer running a simulator; it is choosing among large families of plausible simulators, each corresponding to different forward models/hypotheses consistent with observations. Over large model families, classical Bayesian workflows for model selection are impractical. Furthermore, amortized model selection methods typically hard-code a fixed model prior or complexity penalty at training time, req