EVIA: Entropic Variational Inference Auto-encoding

Yunfei Teng, Zhichao Chen, Xinyu Chen, Lulu Tang, Sixin Zhang, Zhouchen Lin
Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence, PMLR 337:6701-6717, 2026.

Abstract

We present Entropic Variational Inference Auto-encoding ({EVIA}), a novel framework that extends classical variational approaches—including WAEs and {AAEs}, by incorporating entropy regularization. This regularization yields an entropic objective characterized by a closed-form {Gibbs} posterior, established via the {Donsker}–{Varadhan} representation. Consequently, this formulation facilitates efficient sampling strategy, allowing the model to perform high-fidelity autoencoding while jointly ensuring accurate distributional alignment. Empirical results demonstrate {EVIA}’s strong, consistent performance across diverse generative modeling tasks, including posterior estimation, variational inference, and image inpainting.

Cite this Paper

BibTeX
@InProceedings{pmlr-v337-teng26a, title = {{EVIA}: Entropic Variational Inference Auto-encoding}, author = {Teng, Yunfei and Chen, Zhichao and Chen, Xinyu and Tang, Lulu and Zhang, Sixin and Lin, Zhouchen}, booktitle = {Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence}, pages = {6701--6717}, year = {2026}, editor = {Perković, Emilija and Malinsky, Daniel}, volume = {337}, series = {Proceedings of Machine Learning Research}, month = {17--21 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v337/main/assets/teng26a/teng26a.pdf}, url = {https://proceedings.mlr.press/v337/teng26a.html}, abstract = {We present Entropic Variational Inference Auto-encoding ({EVIA}), a novel framework that extends classical variational approaches—including WAEs and {AAEs}, by incorporating entropy regularization. This regularization yields an entropic objective characterized by a closed-form {Gibbs} posterior, established via the {Donsker}–{Varadhan} representation. Consequently, this formulation facilitates efficient sampling strategy, allowing the model to perform high-fidelity autoencoding while jointly ensuring accurate distributional alignment. Empirical results demonstrate {EVIA}’s strong, consistent performance across diverse generative modeling tasks, including posterior estimation, variational inference, and image inpainting.} }
Endnote
%0 Conference Paper %T EVIA: Entropic Variational Inference Auto-encoding %A Yunfei Teng %A Zhichao Chen %A Xinyu Chen %A Lulu Tang %A Sixin Zhang %A Zhouchen Lin %B Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence %C Proceedings of Machine Learning Research %D 2026 %E Emilija Perković %E Daniel Malinsky %F pmlr-v337-teng26a %I PMLR %P 6701--6717 %U https://proceedings.mlr.press/v337/teng26a.html %V 337 %X We present Entropic Variational Inference Auto-encoding ({EVIA}), a novel framework that extends classical variational approaches—including WAEs and {AAEs}, by incorporating entropy regularization. This regularization yields an entropic objective characterized by a closed-form {Gibbs} posterior, established via the {Donsker}–{Varadhan} representation. Consequently, this formulation facilitates efficient sampling strategy, allowing the model to perform high-fidelity autoencoding while jointly ensuring accurate distributional alignment. Empirical results demonstrate {EVIA}’s strong, consistent performance across diverse generative modeling tasks, including posterior estimation, variational inference, and image inpainting.
APA
Teng, Y., Chen, Z., Chen, X., Tang, L., Zhang, S. & Lin, Z.. (2026). EVIA: Entropic Variational Inference Auto-encoding. Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence, in Proceedings of Machine Learning Research 337:6701-6717 Available from https://proceedings.mlr.press/v337/teng26a.html.

Related Material