CutClean: Neural Network Pruning for Privacy-Preserving Inference

Excerpt

arXiv:2608.13773v1 Announce Type: cross 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 fl