EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Adaptive Variational Inference: Beyond Bethe, Tree-Reweighted, and Convex Free Energies

Harald Leisenberger, Franz Pernkopf

IEEE Transactions on Pattern Analysis and Machine Intelligence · 2026-10 · 48 (10) · 13767-13774

DOI: 10.1109/tpami.2026.3704719 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026-10. Verfasst von Harald Leisenberger und 1 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Variational inference in probabilistic graphical models aims to approximate fundamental quantities such as marginal distributions and the partition function. Popular approaches are the Bethe approximation, tree-reweighted, and other types of convex free energies. These approximations are efficient but can fail when the model is complex and highly interactive. In this work, we analyze two classes of approximations that generalize the above methods as special cases: one resulting from changing the state energy of the model, and the other from choosing a different entropy approximation. We discuss benefits and limitations of each approximation, identify favorable regimes for their parameters, and propose ADAPT-$c$c and ADAPT-$\zeta$ζ to automatically detect these optimal settings for a given pairwise binary graphical model. In experiments, we demonstrate their effectiveness for approximating marginals and the partition function.

Zitieren

APA

Leisenberger, H., & Pernkopf, F. (2026). Adaptive Variational Inference: Beyond Bethe, Tree-Reweighted, and Convex Free Energies. IEEE Transactions on Pattern Analysis and Machine Intelligence, 48(10), 13767-13774. https://doi.org/10.1109/tpami.2026.3704719

Vancouver

Leisenberger H, Pernkopf F. Adaptive Variational Inference: Beyond Bethe, Tree-Reweighted, and Convex Free Energies. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2026;48(10):13767-13774. doi:10.1109/tpami.2026.3704719

BibTeX
@article{Leisenberger2026,
  author = {Leisenberger, Harald and Pernkopf, Franz},
  title = {Adaptive Variational Inference: Beyond Bethe, Tree-Reweighted, and Convex Free Energies},
  journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year = {2026},
  volume = {48},
  number = {10},
  pages = {13767--13774},
  doi = {10.1109/tpami.2026.3704719}
}
Kostenlos starten →