{"id":10396,"date":"2026-02-11T07:02:33","date_gmt":"2026-02-11T07:02:33","guid":{"rendered":"https:\/\/mailitics.com\/index.php\/2026\/02\/11\/2602-09277\/"},"modified":"2026-02-11T07:02:33","modified_gmt":"2026-02-11T07:02:33","slug":"2602-09277","status":"publish","type":"post","link":"https:\/\/mailitics.com\/index.php\/2026\/02\/11\/2602-09277\/","title":{"rendered":"Mutual Information Collapse Explains Disentanglement Failure in $beta$-VAEs"},"content":{"rendered":"

Mutual Information Collapse Explains Disentanglement Failure in $beta$-VAEs
\n \t

\n
<\/BR>
\n
\n \t

\n
<\/BR><\/p>\n

arXiv:2602.09277v1 Announce Type: new
\nAbstract: The $beta$-VAE is a foundational framework for unsupervised disentanglement, using $beta$ 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 $beta$ 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 $beta > 1$, stationarity-induced dynamics trigger a spectral contraction of the encoder gain, driving latent-factor mutual information to zero. To resolve this, we introduce the $lambdabeta$-VAE, which decouples regularization pressure from informational collapse via an auxiliary $L_2$ reconstruction penalty $lambda$. Extensive experiments on dSprites, Shapes3D, and MPI3D-real confirm that $lambda > 0$ stabilizes disentanglement and restores latent informativeness over a significantly broader range of $beta$, providing a principled theoretical justification for dual-parameter regularization in variational inference backbones.<\/div>\n

\t

\n
<\/BR>
\n Minh Vu, Xiaoliang Wan, Shuangqing Wei
\n \t

\n
<\/BR>
\nGo to original source<\/a>
\n \t

\n
<\/BR><\/p>\n","protected":false},"excerpt":{"rendered":"

Mutual Information Collapse Explains Disentanglement Failure in $beta$-VAEs arXiv:2602.09277v1 Announce Type: new Abstract: The $beta$-VAE is a foundational framework for unsupervised disentanglement, using $beta$ 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 $beta$ and […]<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[62,113,112],"tags":[460,4755,4756],"class_list":["post-10396","post","type-post","status-publish","format-standard","hentry","category-aimldsaimlds","category-cs-lg","category-stat-ml","tag-beta","tag-collapse","tag-disentanglement"],"_links":{"self":[{"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/posts\/10396"}],"collection":[{"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/comments?post=10396"}],"version-history":[{"count":0,"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/posts\/10396\/revisions"}],"wp:attachment":[{"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/media?parent=10396"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/categories?post=10396"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/mailitics.com\/index.php\/wp-json\/wp\/v2\/tags?post=10396"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}