Gurjeet Sangra Singh, Frantzeska Lavda, Giangiacomo Mercatali, Alexandros Kalousis
The paper introduces a novel grey-box method that integrates incomplete physics models into generative models for learning dynamics from observational data without relying on ground-truth physics parameters.
This research presents a new approach to modeling complex systems by combining deep generative models with traditional physics-based models. Typically, deep learning models can learn from data but often ignore the underlying physics, whereas physics-based models are interpretable but may lack completeness. The proposed method, called Variational Grey-Box Dynamics Matching, bridges this gap by incorporating incomplete physics directly into generative models. This allows for learning dynamics from data alone, without needing exact physical parameters, making it both scalable and stable. The method has shown to perform well in experiments, maintaining the interpretability of physics-based models.
