Identifikacija miona u CMS detektoru pri proton-proton sudarima energije Vs=13 TeV metodama strojnog učenja
2020
In this thesis, the identification of objects obtained by simulation of the Drell-Yan process is
made. The reconstructed objects were divided into two classes: the signal object class and the
background object class, which were then stored in a data set. Signal objects are muons most
likely formed in the primary vertex, while background objects are inaccurately reconstructed as
muons. Data analysis at CERN often use so-called cut-based ID methods to identify objects,
therefore this paper considers the possibility of replacing this approach with a machine learning
strategy. Machine learning teaches the model to distinguish objects by training on already
classified data to be able to identify the signal in the experimental data collected by the operation
of the CMS detector. Wanting to mimic this process, the created data set is divided into a training
data set and a test data set. In achieving the goal, the main impediment were the imbalance of
classes which means the number of background objects is much, approximately 14 times, less
than the signal. Two approaches to the problem of data imbalance are discussed. The first
approach is weight balancing which changes the weights of the error function. The second
approach is to balance classes by adding or removing objects from a training data set.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
0
References
0
Citations
NaN
KQI