NCSI: Accelerating the Inference Pipeline for Particles Track Finding

Status	Completed
Mentor Name	Alina Lazar
Mentor's XSEDE Affiliation	Plan to apply for Research Allocation
Mentor Has Been in XSEDE Community	1-2 years
Project Title	Accelerating the Inference Pipeline for Particles Track Finding
Summary	The main goal of this project is to develop and evaluate a deep learning pipeline for reconstructing particle trajectories. These algorithms can then be used by the scientists at CERN to predict the tracks of the particles in the Large Hadron Collider based on their trajectories. With the ever-growing data from scientific experiments, it is imperative to have automatic ways to analyze that data. Specifically, we work with deep learning models including graph neural networks (GNNs) and multilayer perceptron (MLP) and compare them in terms of accuracy and computing time. The integration of the inference pipeline with the ACTS framework will allow scientists to perform a large set of experiments.
Job Description	- implementing the inference pipeline using Python. We plan to implement both CPU and GPU versions. - convert the inference pipeline to C and C++ to integrate it in the ACTS framework. - performing in depth evaluation of the methods implemented. - read and summarize related research papers - perform data analysis of the results - prepare posters and research papers.
Computational Resources	To perform a real evaluation of the developed methods, access to a supercomputer is essential. Given the large size of the existing data sets, it is required to run these methods on multiple CPUs and GPUs.
Contribution to Community	This project will give the students the opportunity to experience working on a real scientific project. It will be hands on training for HPC. Students will learn CUDA programming and distributed computing.
Position Type	Apprentice
Training Plan	During the first couple of weeks students will be train on how to run python and C++ code on HPC. We will start with simple batch jobs and move to running jobs on multiple CPU cores and multiple GPUs.
Student Prerequisites/Conditions/Qualifications	Students need to know how to program in Python and C++. Familiarity with batch jobs, job scripting and job submission is a plus. GPU programming concepts such as Cuda, also a plus.
Duration	Semester
Start Date	01/10/2022
End Date	05/07/2022