Amortized Bayesian Inference on Multilevel Models of Arbitrary Structure

Name: Amortized Bayesian Inference on Multilevel Models of Arbitrary Structure
Uploaded: 2026-10-04T21:29:08.958341+00:00
Duration: 1242 s
Description: The method derives neural network architectures for multilevel models of any structure automatically using Amortized Bayesian Inference (ABI). It transforms a generative model's graph into a complete set of approximators by applying graph expansion and inversion to find factorizations that efficiently amortize over the number of groups, enabling scalable Bayesian inference.

The method derives neural network architectures for multilevel models of any structure automatically using Amortized Bayesian Inference (ABI). It transforms a generative model's graph into a complete set of approximators by applying graph expansion and inversion to find factorizations that efficiently amortize over the number of groups, enabling scalable Bayesian inference.

Open the full episode