Variational Autoencoders (VAEs) are powerful generative models widely used for learning interpretable latent spaces, quantifying uncertainty, and compressing data for downstream generative tasks. VAEs typically rely on diagonal Gaussian posteriors due to computational constraints. Using arguments grounded in differential geometry, we demonstrate inherent limitations in the representational capacity of diagonal covariance VAEs, as illustrated by explicit low-dimensional examples. In response, we show that a regularized variant of the recently introduced Free-form Injective Flow (FIF) can be interpreted as a VAE featuring a highly flexible, implicitly defined posterior. Crucially, this regularization yields a posterior equivalent to a full Gaussian covariance distribution, yet maintains computational costs comparable to standard diagonal covariance VAEs. Experiments on image datasets validate our approach, demonstrating that incorporating full covariance substantially improves model likelihood.


    Zugriff

    Download


    Exportieren, teilen und zitieren



    Titel :

    Beyond Diagonal Covariance: Flexible Posterior VAEs via Free-Form Injective Flows


    Beteiligte:
    Sorrenson, Peter (Autor:in) / Lührs, Lukas (Autor:in) / Olischläger, Hans (Autor:in) / Köthe, Ullrich (Autor:in)

    Erscheinungsdatum :

    2025


    Medientyp :

    Preprint


    Format :

    Elektronische Ressource


    Sprache :

    Englisch




    f-VAEs: Improve VAEs with Conditional Flows

    Su, Jianlin / Wu, Guang | ArXiv | 2018

    Freier Zugriff

    On Posterior Collapse and Encoder Feature Dispersion in Sequence VAEs

    Long, Teng / Cao, Yanshuai / Cheung, Jackie Chi Kit | ArXiv | 2019

    Freier Zugriff

    LEARNING HIERARCHICAL PRIORS IN VAES

    Klushyn, Alexej / Chen, Nutan / Kurle, Richard et al. | TIBKAT | 2020


    Relaxed-Responsibility Hierarchical Discrete VAEs

    Willetts, Matthew / Miscouridou, Xenia / Roberts, Stephen et al. | ArXiv | 2020

    Freier Zugriff

    On Incorporating Inductive Biases into VAEs

    Miao, Ning / Mathieu, Emile / Siddharth, N. et al. | ArXiv | 2021

    Freier Zugriff