Differentially Private Latent Diffusion Models
Primary Area: societal considerations including fairness, safety, privacy
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: differential privacy, latent diffusion models, generative modelling, synthetic image generation
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: Diffusion models (DMs) are widely used for generating high-quality high-
dimensional images in a non-differentially private manner. However, due to the
notoriously slow training process of DMs, applying differential privacy (DP) to
the training routine requires adding large amounts of noise, yielding poor-quality
generated images. To address this challenge, recent papers suggest pre-training
DMs with public data, then fine-tuning them with private data using DP-SGD for a
relatively short period. In this paper, we aim to further improve the current state
of DMs with DP by adopting the Latent Diffusion Models (LDMs). LDMs are
equipped with powerful pre-trained autoencoders that map the high-dimensional
pixels into lower-dimensional latent representations, where DMs are then trained
in the latent space, yielding a more efficient and fast training. In our algorithm,
DP-LDMs, rather than fine-tuning the entire DMs, we fine-tune only the attention
modules of LDMs at varying layers with privacy-sensitive data, reducing the num-
ber of trainable parameters by approximately 96% compared to fine-tuning the
entire DMs. The smaller parameter space to fine-tune with DP-SGD helps our algo-
rithm to achieve a new state-of-the-art results in several public-private benchmark
data pairs. Our approach also allows us to generate high-dimensional images (256
by 256) and those conditioned on text prompts with differential privacy, which
have not been attempted yet in DP literature. Our approach provides a promising
direction for training more powerful, yet training-efficient differentially private
DMs that can produce high-quality high-dimensional synthetic images.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 4377
Loading