Using machine learning for particle track identification in the CLAS12 detector

Particle track reconstruction is the most computationally intensive process in nuclear physics experiments. Traditional algorithms use a combinatorial approach that exhaustively tests measurements (“hits”) to identify those form an actual particle trajectory . In this article, we describe development of four machine learning (ML) models assist tracking algorithm by identifying valid candidates from drift chambers. Several types were tested, including: Convolutional Neural Networks (CNN), Multi-Layer Perceptrons (MLP), Extremely Randomized Trees (ERT) and Recurrent (RNN). As result work, MLP network classifier was implemented as part CLAS12 software provide code with recommended candidates. The resulting achieved accuracy greater than 99% resulted end-to-end speedup 35% compared existing algorithms.