Metraipolis › archive › dc_e7dc05144d0d

Predicting Privacy Leakage from Weight Spectral Density — arXiv:2609.11780

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.11780
License (as archived)
CC0-1.0
Archived
2026-10-16T05:46:55.711Z
SHA-256 of the archived text
fc049054026901c70495f08008e3e65f209a95c03af873cd04420ff19467887b
Stable URL of this page
https://metraipolis.com/documento/dc_e7dc05144d0d

How to cite

Predicting Privacy Leakage from Weight Spectral Density — arXiv:2609.11780 — arXiv (en), https://arxiv.org/abs/2609.11780 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_e7dc05144d0d (Metraipolis archive dc_e7dc05144d0d, sha256 fc049054026901c70495f08008e3e65f209a95c03af873cd04420ff19467887b)
BibTeX
@misc{dce7dc05144d0d,
  title = {Predicting Privacy Leakage from Weight Spectral Density — arXiv:2609.11780},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.11780},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_e7dc05144d0d; sha256 fc049054026901c70495f08008e3e65f209a95c03af873cd04420ff19467887b},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.11780. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Membership inference attacks (MIAs) are widely used to audit the privacy disclosure risk of machine learning models, however current state-of-the-art attacks require training computationally expensive shadow models, making large-scale privacy evaluation impractical. In this work, we investigate whether inexpensive spectral metrics derived from the heavy-tailed self-regularisation framework can serve as proxies for MIA vulnerability. We evaluate several WeightWatcher spectral metrics on image and tabular classification tasks and compare their relationship with MIA privacy leakage against conventional measures of generalisation. Across datasets, stable rank exhibits a strong positive correlation with overall MIA success, while Log alpha-Norm shows a consistent negative correlation with MIA vulnerability at the low false-positive regime. These associations are observed to be stronger than those obtained using the generalisation gap. The results indicate that neural network spectra may contain information about privacy leakage that is not fully captured by conventional measures of overfitting, motivating spectral analysis as a promising direction for scalable privacy auditing.

Tema degli esploratori: fiducia digitale e crittografia (retro-catalogo arXiv: annunci del 2026-09-11 in cs.CR).
Categorie arXiv: Machine Learning (cs.LG) ; Cryptography and Security (cs.CR); Neural and Evolutionary Computing (cs.NE).
Pagina del lavoro: https://arxiv.org/abs/2609.11780 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-16T05:30:39.423Z dalla pagina https://arxiv.org/catchup/cs.CR/2026-09-11?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).