Particle Track Reconstruction using Machine Learning
- 0 Collaborators
As detectors at CERN get stronger, the luminosity of particle collision increases. This means more and more particles are generated in each event and the classical combinatorial methods reach their limit in processing for such events. Scientists at CERN have turned to the machine learning solutions. ...learn more
Project status: Published/In Market
Groups
Student Developers for AI
Intel Technologies
AI DevCloud / Xeon,
Intel Python,
MKL
Overview / Usage
Particle Track Reconstruction is now a problem to be solved with Machine Learning. The aim is to reconstruct trajectories of particles left as hits on silicon detectors left when traveling in the a direction after they were created during the collision. The amount of hits in an event is approximately 100,000 and the trajectories are only 1000. The tracks can be interpreted as patterns in the data collected during a collision event. This research is going to be used at CERN LHC’s Atlas experiment from 2025.
Methodology / Approach
I’ve explored currently explored methods exposed in the trackml challenge and also developed some of my own methods as I took the part in the competition. The algorithms I’ve worked on including pair prediction using Logistic Regression, classical ML models like Trees, Gradient Boosted trees, Graph Neural Networks, etc. Same goes for triplet prediction, which follows the basic analogy of prediction of a third hit from a pair of hits. The final reconstruction is done usually by an Outlier density estimation algorithm.
Technologies Used
Python, C++, Pytorch, ROOT, ACTS, Keras