![[2602.17477] Variational Grey-Box Dynamics Matching](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.17477] Variational Grey-Box Dynamics Matching
Summary
The paper presents a novel grey-box method that integrates incomplete physics models into deep generative models, enabling the learning of dynamics from observational data without requiring ground-truth physics parameters.
Why It Matters
This research addresses the limitations of traditional physics-based models and black-box generative models by combining their strengths. It offers a scalable and interpretable solution for modeling complex dynamical systems, which is crucial for advancements in fields like robotics and AI safety.
Key Takeaways
- Introduces a grey-box method that combines physics-based models with generative models.
- Learns dynamics from observational data without needing ground-truth parameters.
- Demonstrates performance on par with or superior to fully data-driven approaches.
- Maintains the interpretability of physics models while addressing scalability issues.
- Adapts the framework for second-order dynamics, expanding its applicability.
Computer Science > Machine Learning arXiv:2602.17477 (cs) [Submitted on 19 Feb 2026] Title:Variational Grey-Box Dynamics Matching Authors:Gurjeet Sangra Singh, Frantzeska Lavda, Giangiacomo Mercatali, Alexandros Kalousis View a PDF of the paper titled Variational Grey-Box Dynamics Matching, by Gurjeet Sangra Singh and 3 other authors View PDF HTML (experimental) Abstract:Deep generative models such as flow matching and diffusion models have shown great potential in learning complex distributions and dynamical systems, but often act as black-boxes, neglecting underlying physics. In contrast, physics-based simulation models described by ODEs/PDEs remain interpretable, but may have missing or unknown terms, unable to fully describe real-world observations. We bridge this gap with a novel grey-box method that integrates incomplete physics models directly into generative models. Our approach learns dynamics from observational trajectories alone, without ground-truth physics parameters, in a simulation-free manner that avoids scalability and stability issues of Neural ODEs. The core of our method lies in modelling a structured variational distribution within the flow matching framework, by using two latent encodings: one to model the missing stochasticity and multi-modal velocity, and a second to encode physics parameters as a latent variable with a physics-informed prior. Furthermore, we present an adaptation of the framework to handle second-order dynamics. Our experiments on ...
