Guided-BFNs: Towards Visualizing and Understanding Bayesian Flow Networks in the Context of Trajectory Planning

ICLR 2025 Conference Submission6152 Authors

Keywords: Bayesian flow networks, trajectory planning, model understanding

Abstract:

Bayesian Flow Networks (BFNs) represent an emerging class of generative models that exhibit promising capabilities in modeling continuous, discretized, and discrete data.
In this paper, we develop Guided-BFNs to integrate BFNs with conditional guidance and gradient guidance to facilitate the effective application of such models in trajectory planning tasks.
Based on our developments, we can better comprehend BFNs by inspecting the generation dynamics of the planning trajectories.
Through extensive parameter tuning and rigorous ablation experiments, we systematically delineate the functional roles of various parameters and elucidate the pivotal components within the structure of BFNs. Furthermore, we conduct a comparative analysis of the planning results between diffusion models and BFNs, to discern their similarities and differences.
Additionally, we undertake efforts to augment the performance of BFNs, including developing a faster and training-free sampling algorithm for sample generation.
Our objectives encompass not only a comprehensive exploration of BFNs' structural insights but also the enhancement of their practical utility.

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.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

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: 6152