analiza aiqbench
This paper proposes a single neural inference network that approximates Bayesian posteriors across different priors, likelihoods, latent dimensions, and dataset sizes, without retraining for each new model. If the reported results hold up, it could reduce the cost of Bayesian inference compared with NUTS and per-task variational methods. Readers should look closely at how broadly the benchmarks cover model families and at the accuracy trade-offs against exact sampling.
Amortized Factor Inference Networks for Posterior Inference
arXiv:2605.26419v2 Announce Type: replace Abstract: Amortized inference promises fast test-time Bayesian inference, but existing methods typically target a fixed model or a restricted model family. Extending amortization to new model families often requires retraining or test-time fine-tuning. In this paper, we ask: is it possible to build a single inference network capable of generalizing across varying priors, likelihoods, latent dimensions, and observation counts? We introduce Amortized Factor Inference Networks (AFINs), which describe a model as a list of typed prior and likelihood factors and map this description, together with the observations, to a Gaussian posterior approximation through an encode--merge--decode network whose learned parameters do not depend on the latent dimension or the number of observations. Experiments on synthetic and real-data benchmarks show that a single trained AFIN provides competitive posterior approximations at substantially lower test-time cost than NUTS and per-task variational inference. Optional importance sampling can further improve accuracy.
Abstract:Amortized inference promises fast test-time Bayesian inference, but existing methods typically target a fixed model or a restricted model family. Extending amortization to new model families often requires retraining or test-time fine-tuning. In this paper, we ask: is it possible to build a single inference network capable of generalizing across varying priors, likelihoods, latent dimensions, and observation counts? We introduce Amortized Factor Inference Networks (AFINs), which describe a model as a list of typed prior and likelihood factors and map this description, together with the observations, to a Gaussian posterior approximation through an encode--merge--decode network whose learned parameters do not depend on the latent dimension or the number of observations. Experiments on synthetic and real-data benchmarks show that a single trained AFIN provides competitive posterior approximations at substantially lower test-time cost than NUTS and per-task variational inference. Optional importance sampling can further improve accuracy.
Machine Learning (cs.LG)
arXiv:2605.26419 [cs.LG]
(or arXiv:2605.26419v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2605.26419 Focus to learn more arXiv-issued DOI via DataCite
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.
Text integral sub licența deschisă a publicației · 236 cuvinte
Subiecte conexe
Amortized Bayesian Inference on Multilevel Models of Arbitrary Structure
arXiv · 1 oct. 2026
Amortized Bayesian Causal Discovery of Extended Factor Graphs
arXiv · 28 iul. 2026
Deep Adaptive Bayesian Screening
arXiv · 21 iul. 2026
Prior-Amortized In-Context Bayesian Inference for Generalized Linear Mixed-Effects Models
arXiv · 22 sept. 2026