PRIMS presented at ECML PKDD 2026

Hai-Long Nguyen presented his work, “PRIMS: Physics-guided Representation for Fluid Identification in Multimodal Sensing,” at ECML PKDD 2026, held in Naples, Italy, from 7–11 September.

PRIMS introduces a physics-guided approach that incorporates a fluid-dynamics law directly into the Transformer architecture for multimodal sensing. Evaluated on data from a real micro-fabricated sensor chip, PRIMS achieves 98.9% F1-score for fluid identification while using approximately 14× fewer parameters, and remains robust under previously unseen operating conditions.

The work demonstrates how integrating physical knowledge into machine-learning architectures can lead to models that are both efficient and robust for real-world sensing applications. This research is supported by the NWO MOSAIC project.

Congratulations, Hai-Long, on this amazing work!