We are the Analytics and AI group at SINTEF Digital, based in Oslo. We work on machine learning and data-driven modelling across scientific computing, time-series forecasting, and applied AI for industry. We are part of SINTEF Digital's Department of Mathematics and Cybernetics, on GitHub at @sintefmath. Visit our group webpage for more about who we are and what we do.
| Project | Description |
|---|---|
| Pseudo-Hamiltonian Neural Networks | The phlearn package for modelling pseudo-Hamiltonian systems with neural networks (PHNN), for both ODEs and PDEs. Installable with pip install phlearn. Papers: Eidnes et al., 2023 (Physica D) (arXiv), Eidnes & Lye, 2024 (J. Comput. Phys.) (arXiv). |
| SafetyCage | Python package for detecting misclassified predictions from machine learning models via a unified interface to post-hoc statistical methods (MSP, DOCTOR, Mahalanobis, SPARDACUS, RED). Supports defining custom detection methods. Installable with pip install safetycage. |
| DON-LSTM | An architecture combining DeepONets and LSTMs for multi-resolution learning of dynamical systems, giving accurate long-time predictions of PDEs while needing fewer high-resolution samples. Paper: Michałowska et al., 2025 (Neurocomputing) (arXiv). |
| ReWTS | A Recency-Weighted Temporally-Segmented ensemble for multi-step time-series forecasting, designed to adapt to evolving dynamics in process-industry data. Paper: Johnsen et al., 2025 (JAIR). |
| Road Graph Generator | A tool for mapping roads at construction sites from GPS data, generating road graphs from collected location data. Paper: Michałowska et al., 2025 (Knowledge-Based Systems) (arXiv). |
| Robust ML | Machine learning methods and an evaluation framework for building models that stay reliable under noise, outliers, and distribution shift. |
| NLDL Tutorial | Hands-on tutorial from the Northern Lights Deep Learning (NLDL) 2025 Winter School in Tromsø, on structure-preserving machine learning for physical systems. Three Google Colab notebooks covering physics-informed neural networks (PINNs), pseudo-Hamiltonian neural networks (PHNNs), and Hamiltonian neural networks (HNNs). |
For questions about a specific project, please open an issue or discussion on that project's repository. For general enquiries, contact Signe Riemer-Sørensen, Research Manager for Analytics and AI. Contact details for other researchers in the group can be found on our group webpage.