Latent Anomaly Detection Through Density Matrices

Joseph A. Gallego‑Mejia, Oscar A. Bustos‑Brinez & Fabio A. González
arXiv preprint arXiv:2408.07623 (submitted 15 Aug 2024) (arxiv.org)

Abstract:
This paper introduces a novel anomaly detection framework that combines the robust statistical principles of density‑estimation‑based methods with the representation‑learning power of deep learning. Two versions of the framework are presented: ADDM (Anomaly Detection Through Density Matrices), which uses adaptive Fourier features and explicit density matrix estimation, and LADDM (Latent Anomaly Detection Through Density Matrices), which incorporates a deep autoencoder to learn a low‑dimensional representation and estimate densities in latent space. Extensive experiments across benchmark datasets show that both versions achieve comparable or superior performance to state‑of‑the‑art anomaly detectors, with LADDM especially effective on high‑dimensional data. The methods are end‑to‑end trainable and seamlessly integrate with gradient‑based optimization techniques.
(Keywords: anomaly detection, density matrix, random Fourier features, kernel density estimation, quantum machine learning, deep learning.) :contentReference[oaicite:0]{index=0}

Links:

• 📄 arXiv abstract page
• 📑 Download PDF

📌 BibTeX citation

@article{gallego­mejia2024latent,
  title   = {Latent Anomaly Detection Through Density Matrices},
  author  = {Gallego-Mejia, Joseph A. and Bustos-Brinez, Oscar A. and Gonz\'alez, Fabio A.},
  journal = {arXiv preprint},
  year    = {2024},
  eprint  = {2408.07623},
  archivePrefix = {arXiv},
  primaryClass = {cs.LG},
  url     = {https://arxiv.org/abs/2408.07623}
}