Efficient Diversity-Preserving Diffusion Alignment via Gradient-Informed GFlowNets
Keywords: Reward Finetuning, Diffusion Models, Alignment, GFlowNet, Generative Models
TL;DR: Efficient, diversity-preserving and prior-preserving reward finetuning and alignment of large-scale continuous diffusion models, with a novel generative flow network (GFlowNet) objective that leverages reward gradient signals
Abstract: While one commonly trains large diffusion models by collecting datasets on target downstream tasks, it is often desired to align and finetune pretrained diffusion models with some reward functions that are either designed by experts or learned from small-scale datasets. Existing post-training methods for reward finetuning of diffusion models typically suffer from lack of diversity in generated samples, lack of prior preservation, and/or slow convergence in finetuning. Inspired by recent successes in generative flow networks (GFlowNets), a class of probabilistic models that sample with the unnormalized density of a reward function, we propose a novel GFlowNet method dubbed Nabla-GFlowNet (abbreviated as \nabla-GFlowNet), the first GFlowNet method that leverages the rich signal in reward gradients, together with an objective called \nabla-DB plus its variant residual \nabla-DB designed for prior-preserving diffusion finetuning. We show that our proposed method achieves fast yet diversity- and prior-preserving finetuning of Stable Diffusion, a large-scale text-conditioned image diffusion model, on different realistic reward functions.
Primary Area: generative models
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.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
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: 8475
Loading