Speaker
Description
We present the first application of graph neural networks using GravNet and object condensation to particle flow reconstruction at the International Linear Collider. By embedding both calorimeter hits and tracking information into a latent space, our deep learning model performs simultaneous particle clustering and energy regression. Charged tracks are treated as condensation anchors, improving the separation of overlapping showers. Evaluated on ILD full simulation with tau and light-quark jet samples, the model outperforms PandoraPFA in clustering efficiency and purity. In boosted topologies, where multiple decay products are densely collimated, reducing confusion in particle flow reconstruction is essential. Our approach shows a marked reduction in clustering ambiguities, offering a promising path forward for high-fidelity reconstruction in high-energy jet environments.