Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Stable Port-Hamiltonian Neural Networks
Authors: Fabian J. Roth, Dominik K. Klein, Maximilian Kannapinn, Jan Peters, Oliver Weeger
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Through illustrative and real-world examples, we demonstrate that these strong inductive biases facilitate robust learning of stable dynamics from sparse data, while avoiding instability and surpassing purely data-driven approaches in accuracy and physically meaningful generalization. |
| Researcher Affiliation | Academia | Technical University of Darmstadt, Darmstadt, Germany EMAIL, EMAIL |
| Pseudocode | No | The paper describes the model architecture and mathematical formulations, but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks, nor structured steps formatted like code. |
| Open Source Code | Yes | Code and model weights are available on Git Hub1. 1https://github.com/CPShub/sphnn-publication |
| Open Datasets | Yes | Next, the s PHNN model is evaluated using the cascaded tanks dataset [47, 48]. It contains measurements of a physical fluid level control system consisting of a pump and two tanks, see Appendix D for details. |
| Dataset Splits | Yes | The training data consists of trajectories 745 and 795. They use an amplitude-modulated pseudo-random binary sequence (APRBS) and a multi-sine as the excitation signal Toven... In the experiment, the test group AP15 [21] is utilized as test data. It comprises 15 trajectories with APRBS excitations, specifically selected to ensure an even distribution of amplitudes and median values. |
| Hardware Specification | Yes | The models were trained on a Windows machine with Intel i7-13700K (13th Generation) and 32 GB of memory. |
| Software Dependencies | No | The paper mentions the use of the ADAM optimizer [25] and a numerical integrator using the Runge-Kutta scheme Tsit5, but does not provide specific version numbers for these or other software dependencies like programming languages or deep learning frameworks. |
| Experiment Setup | Yes | Model hyperparameters and computational costs are listed in Appendix B. In each experiment presented within this manuscript, the different models were trained for the same number of steps using identical learning rates. The mean squared error loss function and the ADAM [25] optimizer were used throughout. The model and training hyperparameters, along with the number of model instances trained for statistical evaluation, are provided in Table 2. |