Speaker
Description
Future electron-positron Higgs factories, such as the International Linear Collider (ILC) and FCC-ee, demand unprecedented precision in jet energy measurements to enable detailed studies of the Higgs boson. Highly granular calorimeters coupled with Particle Flow Algorithms (PFA) are central to achieving this performance by leveraging tracking detectors to measure charged particle energies. While PandoraPFA serves as the benchmark algorithm for detector concepts like the International Large Detector (ILD), its performance degrades at jet energies above $100\text{ GeV}$ due to increasing shower overlap and the complexity of track-cluster matching. To address these limitations, we present a machine-learning-based PFA using Graph Neural Networks (GNNs). Our framework adapts the GravNet architecture and the Object Condensation loss function—originally developed for clustering in the CMS High-Granularity Calorimeter (HGCAL)—and extends them to full-detector particle flow by integrating track-cluster matching. In this talk, we present the evaluation of this GNN-based PFA using full ILD detector simulations for $\tau$ leptons and hadronic jet events. We assess clustering efficiency, particle reconstruction capability, and jet energy resolution, highlighting comparative results against PandoraPFA.