Saved in:
Bibliographic Details
Main Authors: Vu, Minh, Wan, Xiaoliang, Wei, Shuangqing
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2602.09277
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914318313848832
author Vu, Minh
Wan, Xiaoliang
Wei, Shuangqing
author_facet Vu, Minh
Wan, Xiaoliang
Wei, Shuangqing
contents The $β$-VAE is a foundational framework for unsupervised disentanglement, using $β$ to regulate the trade-off between latent factorization and reconstruction fidelity. Empirically, however, disentanglement performance exhibits a pervasive non-monotonic trend: benchmarks such as MIG and SAP typically peak at intermediate $β$ and collapse as regularization increases. We demonstrate that this collapse is a fundamental information-theoretic failure, where strong Kullback-Leibler pressure promotes marginal independence at the expense of the latent channel's semantic informativeness. By formalizing this mechanism in a linear-Gaussian setting, we prove that for $β> 1$, stationarity-induced dynamics trigger a spectral contraction of the encoder gain, driving latent-factor mutual information to zero. To resolve this, we introduce the $λβ$-VAE, which decouples regularization pressure from informational collapse via an auxiliary $L_2$ reconstruction penalty $λ$. Extensive experiments on dSprites, Shapes3D, and MPI3D-real confirm that $λ> 0$ stabilizes disentanglement and restores latent informativeness over a significantly broader range of $β$, providing a principled theoretical justification for dual-parameter regularization in variational inference backbones.
format Preprint
id arxiv_https___arxiv_org_abs_2602_09277
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Mutual Information Collapse Explains Disentanglement Failure in $β$-VAEs
Vu, Minh
Wan, Xiaoliang
Wei, Shuangqing
Machine Learning
The $β$-VAE is a foundational framework for unsupervised disentanglement, using $β$ to regulate the trade-off between latent factorization and reconstruction fidelity. Empirically, however, disentanglement performance exhibits a pervasive non-monotonic trend: benchmarks such as MIG and SAP typically peak at intermediate $β$ and collapse as regularization increases. We demonstrate that this collapse is a fundamental information-theoretic failure, where strong Kullback-Leibler pressure promotes marginal independence at the expense of the latent channel's semantic informativeness. By formalizing this mechanism in a linear-Gaussian setting, we prove that for $β> 1$, stationarity-induced dynamics trigger a spectral contraction of the encoder gain, driving latent-factor mutual information to zero. To resolve this, we introduce the $λβ$-VAE, which decouples regularization pressure from informational collapse via an auxiliary $L_2$ reconstruction penalty $λ$. Extensive experiments on dSprites, Shapes3D, and MPI3D-real confirm that $λ> 0$ stabilizes disentanglement and restores latent informativeness over a significantly broader range of $β$, providing a principled theoretical justification for dual-parameter regularization in variational inference backbones.
title Mutual Information Collapse Explains Disentanglement Failure in $β$-VAEs
topic Machine Learning
url https://arxiv.org/abs/2602.09277