Projects
Privacy-Preserving Generative Models via Latent Diffusion
A differentially private autoencoder and diffusion model pipeline. Trained on CIFAR-10 with Opacus for strict noise accounting. Demonstrates high utility despite privacy constraints.
A differentially private autoencoder and diffusion model pipeline. Trained on CIFAR-10 with Opacus for strict noise accounting. Demonstrates high utility despite privacy constraints.