Track Finding in Real and Fake Data using Machine Learning

Future Electron-Ion Collider experiments will have high rate running conditions, and it will be necessary to make quick on the fly decisions about track reconstructions. Machine learning can be used to analyze data in real time and help make these decisions. We researched techniques and tools currently in use, specifically in the LHCb experiment for rejecting fake data, Google’s TensorFlow, and the Keras TensorFlow API. We created models for data from a Hall C experiment at Jefferson Lab. An accuracy of roughly 70% was achieved by training convolutional neural networks on the data. In an effort to create data that could be easily manipulated, a program was made that creates points in 3-D space, some belonging to a track, and some being noise hits. The next steps will be to use both convolutional and recurrent neural networks to find tracks, both from real and fake data.