Speaker
Description
Particle identification (PID) is one of the mostly commonly used tools for the physics analysis in collider physics experiments. To achieve good PID performance, information given by multiple sub-detectors are usually combined. This is in particular necessary for the discrimination of charged particles that have close masses (e.g. muon and pion). However, due to the intrinsic correlations between the input variables, it is usually difficult to apply traditional methods such as the maximum likelihood method.
In the past decades, with the rapid development of machine learning (ML) techniques, lots of successful applications emerged in HEP experiments. One of the obvious advantages of applying ML to PID is its capability of combing many correlated variables with the data-driven approach.
In this work, targeting at the muon/pion identification problem at the BESIII experiment, a electron-positron collider experiment working in the τ−charm energy region, we have developed a muon/pion classifier based on the gradient boosted decision tree (BDT) algorithm. Preliminary results show that the BDT model provides obviously higher discrimination power than traditional methods. In addition, based on the substantial amount of high-quality data taken by the BESIII detector, the method of evaluating and suppressing the systematical error of the ML model will be introduced as well, which is critical for applying the model to physics studies.
| Consider for long presentation | No |
|---|
