aiqbench analysis
This research proposes using lightweight spectral metrics derived from weight distributions as efficient proxies for detecting membership inference attacks, bypassing the need for computationally expensive shadow models. By demonstrating that stable rank and Log alpha-Norm correlate more strongly with privacy leakage than traditional generalization gaps, the study offers a scalable method for auditing model vulnerability without sacrificing accuracy. This approach addresses a critical bottleneck in large-scale privacy evaluation by enabling faster, resource-efficient security assessments for machine learning systems.
Predicting Privacy Leakage from Weight Spectral Density
arXiv:2609.11780v1 Announce Type: new Abstract: 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.
Abstract: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.
Machine Learning (cs.LG); Cryptography and Security (cs.CR); Neural and Evolutionary Computing (cs.NE)
arXiv:2609.11780 [cs.LG]
(or arXiv:2609.11780v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2609.11780 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 259 words
Related coverage
Operator-informed initialization for Fourier features physics-informed neural networks
arXiv · Oct 5, 2026
When Can a Cavity Move a Mott Transition? A Spectral-Density Criterion within Gutzwiller Theory
arXiv · Jul 27, 2026
Automating Learner Assessment: Benchmarking Machine Learning and Deep Learning Models for EEG-Based Familiarity Prediction
arXiv · Aug 19, 2026
Linking Scalar-Intensity Language to Structural Polarization with Validated Signed-Network Measures
arXiv · Oct 6, 2026