Artificial Neural Network Approach to Detector Configuration Optimization Based on the Impact Parameter Estimation Problem

In our work we investigated the application of artificial neural networks (ANNs) to event-wise analysis of heavy ion collision data. We focused on solving the problem of impact parameter estimation using simulated data from a microchannel plate detector (MCP) for potential use in NICA (Nuclotron-based Ion Collider fAcility) collider experiments. Our study reveals that such a technique can be utilized to estimate the impact parameter quite accurately from raw detector data based on the QGSM event generator, specifically from spatial distributions of particles and time-of-flight distributions. However, ANNs results are highly dependent on the model of event generator used to create the dataset. Repeating the experiments with data from an alternative generator based on the EPOS collision model yielded different results. Despite this model dependence of the ANNs, we discuss the way they can be used for extraction of model-independent information. Moreover, we have shown that the detector parameters providing the best reconstruction of the event parameters do not depend on the Monte Carlo model of the event and, therefore, are more likely to be optimal in future collider experiments.