Timezone: »

 
Differentially Private Generation of High Fidelity Samples From Diffusion Models
Vikash Sehwag · Ashwinee Panda · Ashwini Pokle · Xinyu Tang · Saeed Mahloujifar · Mung Chiang · Zico Kolter · Prateek Mittal

@
in Challenges in Deployable Generative AI »
Event URL: https://openreview.net/forum?id=vuVGcl0ed1 »

Diffusion based generative models achieve unprecedented image quality but are known to leak private information about the training data. Our goal is to provide provable guarantees on privacy leakage of training data while simultaneously enabling generation of high-fidelity samples. Our proposed approach first non-privately trains an ensemble of diffusion models and then aggregates their prediction to provide privacy guarantees for generated samples. We demonstrate the success of our approach on the MNIST and CIFAR-10.

Keywords: memorization · Diffusion Models · synthetic data · Generative AI · differential privacy · Privacy leakage

Author Information

Vikash Sehwag (Princeton University)
Ashwinee Panda (Princeton University)
Ashwini Pokle (Carnegie Mellon University)
Xinyu Tang (Princeton University)
Saeed Mahloujifar (Meta)
Mung Chiang (Purdue University)
Zico Kolter (Carnegie Mellon University / Bosch Center for AI)
Prateek Mittal (Princeton University)

More from the Same Authors