Machine Learning x Particle Physics

Building better machine learning tools for physics

ML x Particle Physics, Graph Neural Nets

I’ve also developed strategies using the cornerstone of artificial intelligence to advance the natural sciences. I used to work on graph-based approaches to particle track reconstruction (similar to the TrackML Challenge on Kaggle) - specifically using the representation of 3D point cloud data as a (lower-dimension) graph followed by training a graph neural network on it, possibly conditioned on additional physical information (meta data). Problems in high-energy physics and science in general prove to be a rich testbed for statistical machine learning and Bayesian inference. It is exciting to see a growing focus on making this area more practical.

I spent a wonderful ~2 year period on-and-off at CERN, between 2017 and 2019

2018: CERN Technical Student

  • I won the Google Cloud Award at the Deep Learning Indaba for our work on ML x Particle Physics using DeepJet
  • Preprint of our work on machine learning for high-energy physics
  • Selected as one of the youngest attendees and presented DeepJet at the ML Summer School, Madrid
  • Placed 2nd in the CodaLab Challenge at the ML for High-energy Physics School organised by Yandex at Oxford University
  • I was one of the youngest finalists for the $150,000 Reliance Dhirubhai Scholarship for pursuing an MBA degree at Stanford University (declined)
  • Presented DeepJet at the CERN IML Workshop and other Working Group Meetings at CERN

Summer Schools

Podcast
  • I participated in my first podcast discussing CERN, higher-education, and more with some fantastic students from my alma mater

Talks

I have given a set of talks on topics including internships, working in quantum physics (from a Computer Scientist’s perspective), my work at CERN on deep learning for jet physics, and a project presentation from my work at IIT Bombay and Microsoft Research.

CERN: The DeepJet Framework

My primary project at CERN: Build and deploy a Python package for training and evaluation of deep neural networks for “jet” tagging in high-energy physics.

  1. Talk at the Machine Learning Working Group Meeting, April 2018

  2. Talk at the CMS Machine Learning Workshop, July 2018

  3. Invited talk for the CVIT Lab, Indian Institute of Information Technology, Hyderabad:

Please note the references for the CVIT (IIIT-H) talk in the last slide of the presentation titled ‘The DeepJet Framework.’ Some of the slides used are (as cited) from various presentations for the public built by different researchers at the CMS Experiment.