Tracknetv3 With Optimized Inference for Bm@n Tracking

Abstract

Tracking is an important task in the field of High Energy physics. Modern experiments produceenormous amounts of data, and classical tracking algorithms cannot reach required computingefficiency. This lead to the need to develop new methods, some of them use neural network models.In our work we present modifications of previously developed model, TrackNetV2. This model and itsdescendants showed great results for Monte-Carlo simulations of experiments with microstrip-basedGEM detectors: BESIII and BM@N RUN6. In this work we adapt it to more complex scenario forBM@N RUN7. The work showed limitations in architecture and training procedure, which arereworked later.

Deep neural network applications for particle tracking at the BM@N and SPD experiments

“…Several deep neural networks were used to perform tracking at BM@N [2,3]. Those models may be classified into local and global approaches depending on what part of data is used during the inference phase.…”

“…Those models may be classified into local and global approaches depending on what part of data is used during the inference phase. TrackNETv3 [2] sees only one particular track at time, so it can be considered as a local model. On the contrary, the GraphNet model [3] operates globally, using all hits in the event at once.…”

“…The base training procedure of TrackNETv3 is described in detail in [2]. Detector insufficiency was considered too.…”

“…The neural network approach has been successfully implemented in some works devoted to particle tracking in modern experiments. For example, graphs (GNNs) and recurrent neural networks (RNNs) have shown high efficiency for Monte Carlo simulated data of the BESIII and BM@N experiments [2,3].…”