Pattern recognition algorithm for the reconstruction of particles published in MLST!

Our postdoc Will Sutcliffe and collaborators have developed a novel Heterogeneous Graph Neural Network (HGNN) architecture that tackles one of the most demanding real-time data processing challenges in modern physics - reconstructing particle collision events at the Large Hadron Collider. By combining multi-task learning with intelligent graph pruning, the system simultaneously performs multiple complex classification tasks within a single scalable framework, achieving significant improvements in both accuracy and inference speed. This work demonstrates how cutting-edge machine learning techniques can be applied to high-throughput, low-latency environments where data volumes and complexity are extreme - challenges that are equally relevant across industries such as telecommunications, autonomous systems, financial technology, and large-scale sensor networks. Published in Machine Learning: Science and Technology.