EUVIMEDEUVIMED entdecken
Crossref

Information-Geometric Expressivity Diagnostics for Quantum Molecular Autoencoders via Fisher Effective Rank: A Theoretical Study

Ziwei Ma, Yingfeng Wang

Proceedings of the AAAI Symposium Series · 06.11.2026 · 10 (1) · 787-793

DOI: 10.1609/aaaiss.v10i1.43490 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the AAAI Symposium Series, online erschienen am 06.11.2026. Verfasst von Ziwei Ma und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

We develop an information-geometric theory of expressivity for variational quantum models with binary readout--parameterized circuits whose measurement outcome is dichotomic, so that the induced likelihood is Bernoulli. This class includes quantum autoencoders, fidelity and SWAP-test estimators, and binary variational classifiers. For any model in the class we derive the Classical Fisher Information Matrix (CFIM) in closed form and study the Effective Rank k--a spectral-entropy summary of the CFIM eigenspectrum--as a data-aware expressivity measure. We prove four results, each stated for the general class. First, a Data-Limited Rank Bound on the algebraic CFIM rank. Second, a spectral inequality k

Zitieren

APA

Ma, Z., & Wang, Y. (2026). Information-Geometric Expressivity Diagnostics for Quantum Molecular Autoencoders via Fisher Effective Rank: A Theoretical Study. Proceedings of the AAAI Symposium Series, 10(1), 787-793. https://doi.org/10.1609/aaaiss.v10i1.43490

Vancouver

Ma Z, Wang Y. Information-Geometric Expressivity Diagnostics for Quantum Molecular Autoencoders via Fisher Effective Rank: A Theoretical Study. Proceedings of the AAAI Symposium Series. 2026;10(1):787-793. doi:10.1609/aaaiss.v10i1.43490

BibTeX
@article{Ma2026,
  author = {Ma, Ziwei and Wang, Yingfeng},
  title = {Information-Geometric Expressivity Diagnostics for Quantum Molecular Autoencoders via Fisher Effective Rank: A Theoretical Study},
  journal = {Proceedings of the AAAI Symposium Series},
  year = {2026},
  volume = {10},
  number = {1},
  pages = {787--793},
  doi = {10.1609/aaaiss.v10i1.43490}
}
Kostenlos starten →