aiqbench analysis
This research introduces Privacy-aware Sparsity Tuning (PAST), a method that selectively penalizes model parameters most responsible for membership inference vulnerabilities rather than applying uniform regularization. By targeting specific high-risk weights, PAST effectively narrows the loss gap between training and non-training samples without significantly degrading overall model utility. The approach offers a practical pathway to strengthening privacy guarantees in over-parameterized deep learning models against sophisticated inference attacks.
Defending Membership Inference Attacks via Privacy-aware Sparsity Tuning
arXiv:2410.06814v2 Announce Type: replace-cross Abstract: Over-parameterized models are typically vulnerable to membership inference attacks, which aim to determine whether a specific sample is included in the training of a given model. Previous Weight regularizations (e.g., L1 regularization) typically impose uniform penalties on all parameters, leading to a suboptimal tradeoff between model utility and privacy. In this work, we first show that only a small fraction of parameters substantially impact the privacy risk. In light of this, we propose Privacy-aware Sparsity Tuning (PAST), a simple fix to the L1 Regularization, by employing adaptive penalties to different parameters. Our key idea behind PAST is to promote sparsity in parameters that significantly contribute to privacy leakage. In particular, we construct the adaptive weight for each parameter based on its privacy sensitivity, i.e., the gradient of the loss gap with respect to the parameter. Using PAST, the network shrinks the loss gap between members and non-members, leading to strong resistance to privacy attacks. Extensive experiments demonstrate the superiority of PAST, achieving a state-of-the-art balance in the privacy-utility trade-off.
Abstract:Over-parameterized models are typically vulnerable to membership inference attacks, which aim to determine whether a specific sample is included in the training of a given model. Previous Weight regularizations (e.g., L1 regularization) typically impose uniform penalties on all parameters, leading to a suboptimal tradeoff between model utility and privacy. In this work, we first show that only a small fraction of parameters substantially impact the privacy risk. In light of this, we propose Privacy-aware Sparsity Tuning (PAST), a simple fix to the L1 Regularization, by employing adaptive penalties to different parameters. Our key idea behind PAST is to promote sparsity in parameters that significantly contribute to privacy leakage. In particular, we construct the adaptive weight for each parameter based on its privacy sensitivity, i.e., the gradient of the loss gap with respect to the parameter. Using PAST, the network shrinks the loss gap between members and non-members, leading to strong resistance to privacy attacks. Extensive experiments demonstrate the superiority of PAST, achieving a state-of-the-art balance in the privacy-utility trade-off.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2410.06814 [cs.LG]
(or arXiv:2410.06814v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2410.06814 Focus to learn more arXiv-issued DOI via DataCite
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 · 264 words
Related coverage
FISGuard: Defending Against Membership Inference via Fixed Input Subspaces
arXiv · Sep 1, 2026
On the Privacy of LLMs: An Ablation Study
arXiv · Oct 17, 2026
Research Note: Filtering Subversion-Relevant Information From Pretraining Data Is Feasible
LessWrong · Oct 5, 2026
LiveLaw Supreme Court Half-Yearly Digest: January - June 2026 [Part 2]
livelaw.in · Sep 13, 2026