aiqbench analysis
CutClean introduces a novel pruning technique using auxiliary privacy heads to quantify and suppress information leakage in neural networks without relying on traditional dataset bias metrics. By increasing model sparsity while maintaining classification accuracy, this method directly addresses critical security concerns for deploying AI systems handling sensitive attributes. The research demonstrates that significant private data flow can be mitigated through structural network modification rather than just input data curation.
CutClean: Neural Network Pruning for Privacy-Preserving Inference
arXiv:2608.13773v1 Announce Type: new Abstract: Neural networks are increasingly deployed in high-stakes applications with growing privacy leakage concerns. We show that this privacy leakage can occur even in the absence of representation imbalances that lead to traditional dataset biases. This poses significant privacy risks when deploying models that process sensitive attributes. In this context, we propose CutClean, a privacy-aware pruning method that allows to reduce privacy information flow through the network, while increasing its sparsity. Our approach employs auxiliary linear privacy heads placed at each network's block to quantify information leakage, and further applies increasing levels of sparsity to remove the private attribute leakage, measured in terms of the accuracy of the privacy head attached to the last block. Experiments on synthetic and real-world datasets demonstrate that our approach effectively minimizes private information flow while achieving high sparsity rates and preserving classification target accuracy.
Abstract:Neural networks are increasingly deployed in high-stakes applications with growing privacy leakage concerns. We show that this privacy leakage can occur even in the absence of representation imbalances that lead to traditional dataset biases. This poses significant privacy risks when deploying models that process sensitive attributes. In this context, we propose CutClean, a privacy-aware pruning method that allows to reduce privacy information flow through the network, while increasing its sparsity. Our approach employs auxiliary linear privacy heads placed at each network's block to quantify information leakage, and further applies increasing levels of sparsity to remove the private attribute leakage, measured in terms of the accuracy of the privacy head attached to the last block. Experiments on synthetic and real-world datasets demonstrate that our approach effectively minimizes private information flow while achieving high sparsity rates and preserving classification target accuracy.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2608.13773 [cs.LG]
(or arXiv:2608.13773v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2608.13773 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
In: De Marsico, M., et al. Pattern Recognition. ICPR 2026. Lecture Notes in Computer Science, vol 16822. Springer, Cham
https://doi.org/10.1007/978-3-032-31452-9_30 Focus to learn more DOI(s) linking to related resources
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 · 265 words
Related coverage
Efficient EEG Seizure Detection Using INT8 Quantization, Channel Pruning, and Spiking Neural Networks
arXiv · Jul 21, 2026
Reliable Federated TinyML Deployment for IoT Security
arXiv · Sep 24, 2026
Loss-Aware Feature-Map Pruning in Convolutional Neural Networks Using Multi-Armed Bandits
arXiv · Jul 28, 2026
Exploring the Trade-Off Between Structured Pruning and Fault Tolerance in Deep Neural Networks for Space Applications
arXiv · Oct 5, 2026