P-Align: Self-Alignment in Physical Dynamical System Modeling
Keywords: dynamic systems modeling, physical consistency.
Abstract: Deep learning has emerged as the new paradigm in modeling complex physical dynamical systems. Nevertheless, data-driven methods learn patterns by optimizing statistical metrics, tend to overlook the adherence to physical laws. Previous work have attempted to incorporate physical constraints into neural networks, but they often face limitations due to lack of flexibility or optimization challenges. In this paper, we propose a novel framework, Physics-aware Self-Alignment (P-Align), to enhance the physical consistency of dynamical systems modeling. P-Align enables dynamical system models to provides physics-aware rewards, which makes self-alignment of dynamical system models possible. Comprehensive experiments show that \method{} not only gave an average statistical skill score boost of more than 32% for ten backbones on five datasets, but also significantly enhances physics-aware metrics. All of our source codes will be released via GitHub.
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
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: 3054
Loading