aiqbench analysis
This research introduces a novel framework for amortized Bayesian inference on complex multilevel models, automatically deriving neural network architectures from directed acyclic graphs without simplifying dependency structures. By preserving full conditional independence assumptions, the method achieves near-instant inference speeds even on models exceeding 6,500 parameters, offering a significant efficiency gain over traditional gold-standard samplers.
Amortized Bayesian Inference on Multilevel Models of Arbitrary Structure
arXiv:2609.40024v1 Announce Type: cross Abstract: We develop a general method for amortized Bayesian inference on multilevel models of arbitrary structure. Given a generative model specified as a directed acyclic graph, our method automatically derives valid factorizations of the joint posterior and matching neural network architectures. The key steps, graph expansion and graph inversion, yield an inverse graph that determines how inference networks are stacked and conditioned, producing factorizations that amortize over the number of groups and the number of observations within each group. Unlike approaches that simplify the dependency structure to speed up learning or inference, our method preserves all conditional independence and exchangeability assumptions of the generative model. Across three case studies, it closely matches gold-standard samplers on models with more than 6,500 parameters while reducing inference to a near-instant forward pass once trained.
Abstract:We develop a general method for amortized Bayesian inference on multilevel models of arbitrary structure. Given a generative model specified as a directed acyclic graph, our method automatically derives valid factorizations of the joint posterior and matching neural network architectures. The key steps, graph expansion and graph inversion, yield an inverse graph that determines how inference networks are stacked and conditioned, producing factorizations that amortize over the number of groups and the number of observations within each group. Unlike approaches that simplify the dependency structure to speed up learning or inference, our method preserves all conditional independence and exchangeability assumptions of the generative model. Across three case studies, it closely matches gold-standard samplers on models with more than 6,500 parameters while reducing inference to a near-instant forward pass once trained.
Machine Learning (stat.ML); Machine Learning (cs.LG); Computation (stat.CO)
arXiv:2609.40024 [stat.ML]
(or arXiv:2609.40024v1 [stat.ML] for this version)
https://doi.org/10.48550/arXiv.2609.40024 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 229 words
Related coverage
ProximalFM: Amortized Proximal Causal Inference under Hidden Confounding
arXiv · Oct 7, 2026
Semi-Amortized Bayesian Inference with Reusable Local Posteriors
arXiv · Oct 15, 2026
Hierarchical Latent Structure Learning through Online Inference
arXiv · Jul 27, 2026
Prior-Amortized In-Context Bayesian Inference for Generalized Linear Mixed-Effects Models
arXiv · Sep 22, 2026