Hellinger Multimodal Variational Autoencoders

Huyen Thuc Khanh Vo, Isabel Valera
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:2260-2268, 2026.

Abstract

Multimodal variational autoencoders (VAEs) are widely used for weakly supervised generative learning with multiple modalities. Predominant methods aggregate unimodal inference distributions using either a product of experts (PoE), a mixture of experts (MoE), or their combinations to approximate the joint posterior. In this work, we revisit multimodal inference through the lens of probabilistic opinion pooling, an optimization-based approach. We start from H{ö}lder pooling with $\alpha=0.5$, which corresponds to the unique symmetric member of the $\alpha$-divergence family, and derive a moment-matching approximation, termed Hellinger. We then leverage such an approximation to propose HELVAE, a multimodal VAE that avoids sub-sampling, yielding an efficient yet effective model that: (i) learns more expressive latent representations as additional modalities are observed; and (ii) empirically achieves better trade-offs between generative coherence and quality, outperforming state-of-the-art multimodal VAE models.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-vo26b, title = { Hellinger Multimodal Variational Autoencoders }, author = {Vo, Huyen Thuc Khanh and Valera, Isabel}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {2260--2268}, 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/vo26b/vo26b.pdf}, url = {https://proceedings.mlr.press/v300/vo26b.html}, abstract = { Multimodal variational autoencoders (VAEs) are widely used for weakly supervised generative learning with multiple modalities. Predominant methods aggregate unimodal inference distributions using either a product of experts (PoE), a mixture of experts (MoE), or their combinations to approximate the joint posterior. In this work, we revisit multimodal inference through the lens of probabilistic opinion pooling, an optimization-based approach. We start from H{ö}lder pooling with $\alpha=0.5$, which corresponds to the unique symmetric member of the $\alpha$-divergence family, and derive a moment-matching approximation, termed Hellinger. We then leverage such an approximation to propose HELVAE, a multimodal VAE that avoids sub-sampling, yielding an efficient yet effective model that: (i) learns more expressive latent representations as additional modalities are observed; and (ii) empirically achieves better trade-offs between generative coherence and quality, outperforming state-of-the-art multimodal VAE models. } }
Endnote
%0 Conference Paper %T Hellinger Multimodal Variational Autoencoders %A Huyen Thuc Khanh Vo %A Isabel Valera %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-vo26b %I PMLR %P 2260--2268 %U https://proceedings.mlr.press/v300/vo26b.html %V 300 %X Multimodal variational autoencoders (VAEs) are widely used for weakly supervised generative learning with multiple modalities. Predominant methods aggregate unimodal inference distributions using either a product of experts (PoE), a mixture of experts (MoE), or their combinations to approximate the joint posterior. In this work, we revisit multimodal inference through the lens of probabilistic opinion pooling, an optimization-based approach. We start from H{ö}lder pooling with $\alpha=0.5$, which corresponds to the unique symmetric member of the $\alpha$-divergence family, and derive a moment-matching approximation, termed Hellinger. We then leverage such an approximation to propose HELVAE, a multimodal VAE that avoids sub-sampling, yielding an efficient yet effective model that: (i) learns more expressive latent representations as additional modalities are observed; and (ii) empirically achieves better trade-offs between generative coherence and quality, outperforming state-of-the-art multimodal VAE models.
APA
Vo, H.T.K. & Valera, I.. (2026). Hellinger Multimodal Variational Autoencoders . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:2260-2268 Available from https://proceedings.mlr.press/v300/vo26b.html.

Related Material