Saved in:
Bibliographic Details
Main Authors: Vo, Huyen, Valera, Isabel
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.06572
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912989646422016
author Vo, Huyen
Valera, Isabel
author_facet Vo, Huyen
Valera, Isabel
contents 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 Hölder pooling with $α=0.5$, which corresponds to the unique symmetric member of the $α\text{-divergence}$ 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.
format Preprint
id arxiv_https___arxiv_org_abs_2601_06572
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Hellinger Multimodal Variational Autoencoders
Vo, Huyen
Valera, Isabel
Machine Learning
Artificial Intelligence
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 Hölder pooling with $α=0.5$, which corresponds to the unique symmetric member of the $α\text{-divergence}$ 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.
title Hellinger Multimodal Variational Autoencoders
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2601.06572