GarNet-on-FPGA: Simulated Electron and Pion Hits in an Imaging Calorimeter

Y. Iiyama from The University of Tokyo created a dataset from a simulation of electrons and pions shot at an imaging calorimeter. The dataset contains 10,000 simulated events, each with up to 128 hits featuring x, y, z coordinates and energy, and raw sensor data for 4,375 sensors. It was produced as a case study for GarNet-on-FPGA, documented in arXiv:2008.03601.

Use Cases

• Train particle classification models (electron vs. pion) based on the 'truth_pid' labels and cluster hit features.
• Develop energy regression models based on the 'truth_energy' labels and simulated hit data.
• Benchmark FPGA-accelerated neural network architectures (like GarNet) for processing sparse, irregular detector data.
• Study the performance of algorithms on zero-padded, variable-length sequences based on the 'size' array.

Strengths

• Contains 10,000 simulated events, providing a substantial sample for model training and validation.
• Includes both processed cluster data (128 hits with 4 features) and raw sensor-level data (4,375 sensors with 2 features), offering flexibility for different analysis approaches.
• Provides ground truth labels for particle identity and true energy, enabling supervised learning tasks.

Limitations

• Row count for the primary data arrays is known (10,000), but the overall dataset size in bytes and file formats beyond HDF5 are unknown.
• Last update date is unknown; freshness unverified.
• Data is purely simulated, and its representativeness of real-world detector noise and imperfections may be limited.

Provenance

Source

Y. Iiyama, The University of Tokyo

Collection Method

Output of a simulation of electrons and pions interacting with an imaging calorimeter.