Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
To address extreme pileup in High-Luminosity LHC calorimeters, this work designs an optimized Graph Neural Network (GNN) segmentation block to resolve dense, overlapping two-photon showers. Accurately disentangling these merged energy deposits is critical for high-energy physics reconstruction pipelines that are otherwise constrained by combinatorial latency budgets. By predicting continuous node-level energy fractions via accelerated graph construction and convolution routines, the architecture matches baseline separation efficiency with significantly reduced algorithmic complexity.
Merged particle showers can be cleanly disentangled under extreme collision pileup by regressing node-level energy fractions with an accelerated GNN, delivering high-precision calorimeter segmentation without prohibitive computational complexity.
High-granularity calorimeters at the High-Luminosity LHC require novel algorithms to resolve overlapping particle showers. We present an optimised Graph Neural Network (GNN) segmentation block that predicts node-level energy fractions to disentangle overlapping two-photon showers. By accelerating graph construction and convolution operations, our pipeline achieves competitive separation efficiency with significantly reduced algorithmic complexity.