Project Info
Applying Advance Machine Learning to Ultra-High-Energy Cosmic Ray Particle Showers
Eric Mayotte
emayotte@mines.edu
Project Goals and Description:
This project will continue developing advanced Neural networks aimed at reconstructing the mass of an Ultra-High-Energy Cosmic Ray (UHECR) using simulations of the rate at which they deposit energy in the atmosphere. The precursor MURF of this project has already achieved world record performance on this problem and is rapidly approaching publication. This project will focus on improving the network further and learning how the network achieves this extreme result to learn physics from the trained model itself.
As for why this project is interesting:
UHECR are single atomic nuclei from other galaxies and are the most energetic phenomena known to humankind.
In observing these and using them to increase our astrophysical understanding, two properties are of key importance: the particle's energy and its mass.
When one of these particles strikes the top of the atmosphere, it is destroyed, creating a high-energy particle cascade that can reach 10 billion particles in size.
This project aims to use a new method to maximize the amount of information that can be extracted on the mass of the cosmic ray, which, if very successful, could lead to a revolution in the quality of astrophysics that can be done with current ground-based observatories as well as planned future spaced-based observatories.
The successful MURF student will join an established team consisting of a professor, a postdoc, two Ph.D. students, and one or two other undergraduate researchers. The MURF student will be expected to attend the ML-focused meetings of this group and share their opinion and knowledge with others while also benefiting from their knowledge and experience. The work of each team member will build off of that of the others and publications that arise from this team will be jointly authored when appropriate.
More Information:
Grand Challenge: Engineer the tools of scientific discovery.
The prospective students can familiarize themselves with Ultra-High-Energy Cosmic rays and the Fluorescence detector data they will be using at the Pierre Auger Observatory Outreach page here: https://opendata.auger.org/outreach.php
The prospective students can explore a brief tutorial on using Machine Learning to reconstruct Cosmic Ray data here: https://github.com/jglombitza/tutorial_nn_airshowers
A very thorough overview of the field can be found here: 9207a1e1f4bb7ee13bf0d70389453c01
Primary Contacts:
Eric, Mayotte, emayotte@mines.edu | Sonja, Mayotte, smayotte@mines.edu
Student Preparation
Qualifications
A firm understanding of Python and tools like pyplot and scipy are a must. Any knowledge of ML will be useful. The students will be running their code on a Linux machine so familiarity with using a terminal and ssh will be very useful as well.
TIME COMMITMENT (HRS/WK)
5
SKILLS/TECHNIQUES GAINED
The student will learn how to develop, train, judge, optimize, and test neural networks of various architectures. Depending on the progress, the student will also learn how to write scientific publications and potentially present at conferences.
MENTORING PLAN
The student can expect several forms of advising and mentorship throughout the project.
I will be meeting weekly with the students to discuss research.
The student will gain access to the astrophysics machine learning slack channel where they can reach out to me or other members of the group at anytime to answer questions.
Twice per semester, I will meet with the student to discuss career and academic goals and develop a plan to meet them.
PREFERRED STUDENT STATUS
Junior
Senior