Speaker
Description
Application of machine learning has become highly relevant for solving the problems of event reconstruction, particle identification, and impact parameter estimation [1, 2]. The present study focuses on the application of neural networks to predict the energy released in high-energy collisions of two gold nuclei (Au+Au) based on signals obtained from microchannel plate detectors. The initial data consist of events generated by the QGSM event generator at √(s_NN ) = 11.006 GeV per nucleon–nucleon pair.
For each event, the detector measures the particle’s time-of-flight and hit coordinates on the detector surface, enabling reconstruction of the space-time structure and estimation of the released energy. The pseudorapidity of the detectable particles in the tested configuration is limited to the range 3.5 < η < 5.8. Two neural network architectures were implemented and compared: a fully connected network and a fully connected network with an additional convolutional layer. It has been found that the model with a convolutional layer, where all particles contribute to the released energy in the collision but the feature vector fed into the neural network is formed based on charged particles, gives the best result: root mean square error RMSE = 188.2 GeV and relative error less than 9%. The results confirm the potential of machine learning methods for fast and accurate energy reconstruction in nuclear collision experiments, even with strict geometric constraints.
References
[1] Galaktionov, K. A., Roudnev, V. A., Valiev, F. F. Neural Network Approach to Impact Parameter Estimation in High-Energy Collisions Using Microchannel Plate Detector Data. Moscow University Physics Bulletin, 2023, 78(Suppl 1), S52–S58.
[2] Valiev F. F., Vechernin V. V., Feofilov G. A. Estimation of the Accuracy of Determining the Number of Spectator Nucleons from the Energy Measured in a Calorimeter in A+ A Collisions //Bulletin of the Russian Academy of Sciences: Physics. – 2024. – Т. 88. – №. 8. – С. 1312-1318.
