Reducing Noisy Clusters in the NA61/SHINE Project with Help of Machine Learning

We live in an era where huge amounts of data are being generated in all sectors of science and industry. We call it “Big Data”. With Big data, we face the challenge of its analysis and interpretation. Therefore, the need for novel machine learning and artificial intelligence methods has drastically increased in recent years. Likewise, the range of areas in which machine learning has been successfully applied has increased significantly: including image recognition, speech recognition, natural language processing, computational biology and particle physics. Modern tracking devices for particle collision experiments, such as the four large-volume Time Projection Chambers (TPC) used for the NA61/SHINE project at the European Organization for Nuclear Research (CERN), collect huge amounts of data produced by particle collisions. This data contains not only valuable information but also a lot of noise. As recent breakthroughs inspire us to apply machine learning techniques in more and more areas, the question arises: Can we reduce the noise with help of Machine Learning? In this thesis work, the above mentioned question is answered and multiple machine- and deep learning algorithms are not only proposed, but the underlying principles explained and visualized. This thesis tries to answer why those algorithms work and what they might learn.