Amortized Structural Variational Inference

Shitao Fan, Carlos Misael Madrid Padilla, Yun Yang, Lizhen Lin
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:3700-3708, 2026.

Abstract

Variational inference (VI) is widely used for approximate Bayesian inference, but it can scale poorly and often requires re-optimization when new data arrive. Amortized variational inference (AVI) learns a global inference map, yet standard mean-field AVI can suffer from large variational and amortization gaps because of independence assumptions. We propose amortized structural variational inference (ASVI), which injects structural dependencies among latent variables through neural architectures that encode local neighborhood information. ASVI reduces both gaps while retaining scalability. Simulations and real-data experiments show that ASVI improves predictive accuracy and posterior fidelity over AVI, and matches structured VI at lower computational cost.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-fan26b, title = { Amortized Structural Variational Inference }, author = {Fan, Shitao and Padilla, Carlos Misael Madrid and Yang, Yun and Lin, Lizhen}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {3700--3708}, year = {2026}, editor = {Khan, Emtiyaz and Li, Yingzhen and Solin, Arno and Ramdas, Aaditya}, volume = {300}, series = {Proceedings of Machine Learning Research}, month = {02--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v300/main/assets/fan26b/fan26b.pdf}, url = {https://proceedings.mlr.press/v300/fan26b.html}, abstract = { Variational inference (VI) is widely used for approximate Bayesian inference, but it can scale poorly and often requires re-optimization when new data arrive. Amortized variational inference (AVI) learns a global inference map, yet standard mean-field AVI can suffer from large variational and amortization gaps because of independence assumptions. We propose amortized structural variational inference (ASVI), which injects structural dependencies among latent variables through neural architectures that encode local neighborhood information. ASVI reduces both gaps while retaining scalability. Simulations and real-data experiments show that ASVI improves predictive accuracy and posterior fidelity over AVI, and matches structured VI at lower computational cost. } }
Endnote
%0 Conference Paper %T Amortized Structural Variational Inference %A Shitao Fan %A Carlos Misael Madrid Padilla %A Yun Yang %A Lizhen Lin %B Proceedings of The 29th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2026 %E Emtiyaz Khan %E Yingzhen Li %E Arno Solin %E Aaditya Ramdas %F pmlr-v300-fan26b %I PMLR %P 3700--3708 %U https://proceedings.mlr.press/v300/fan26b.html %V 300 %X Variational inference (VI) is widely used for approximate Bayesian inference, but it can scale poorly and often requires re-optimization when new data arrive. Amortized variational inference (AVI) learns a global inference map, yet standard mean-field AVI can suffer from large variational and amortization gaps because of independence assumptions. We propose amortized structural variational inference (ASVI), which injects structural dependencies among latent variables through neural architectures that encode local neighborhood information. ASVI reduces both gaps while retaining scalability. Simulations and real-data experiments show that ASVI improves predictive accuracy and posterior fidelity over AVI, and matches structured VI at lower computational cost.
APA
Fan, S., Padilla, C.M.M., Yang, Y. & Lin, L.. (2026). Amortized Structural Variational Inference . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:3700-3708 Available from https://proceedings.mlr.press/v300/fan26b.html.

Related Material