Deep Privacy Funnel Model: From a Discriminative to a Generative Approach With an Application to Face Recognition

Summary

This study applies the information-theoretic Privacy Funnel model to face recognition, using an end-to-end trainable framework for privacy-preserving representation learning. It introduces the Generative Privacy Funnel (GenPF) and the deep variational PF (DVPF) model, and reports that the framework yields a controllable privacy-utility trade-off while substantially reducing leakage about sensitive attributes on modern face recognition systems. A PyTorch implementation is released.