This repository implements DFORM, a method to align and compare dynamical systems as well as to identify low-dimensional motifs in high-dimensional systems. The method is described in the following preprint:
Chen, R., Vedovati, G., Braver, T., & Ching, S. (2025). Comparing Dynamical Models Through Diffeomorphic Vector Field Alignment (arXiv:2512.18566). arXiv. https://doi.org/10.48550/arXiv.2512.18566
The main script is DFORM.py which defines the model architecture DFORM and the Training and Testing functions. The hyperparmeters were detailed in the comments of Training function. To train the model with multiple initializations and pick the best one, you can use the get_topo_sim function instead of Training.
example_systems.py defines some example dynamical systems. visualizations.py defines some functions to visualize vector fields: PlotField, PlotVec, PlotTraj and PlotDFORM.