Nisha Lad

I intend to pursue an academic career straddling the interface between fundamental physics and mathematics, guided by modern tools such as machine learning and Bayesian inference.

Project title: Graph Neural Networks for fast track finding in LHC data

Research Group: High Energy Physics 

Supervisor(s): Prof Nikos Konstantinidis

Introduction: 

I graduated with an MSci in Physics from UCL. My Master's project confirmed that the 2D XY-Model of quench dynamics can be realised in a non-equilibrium complex condensed matter system. I investigated this by studying the properties of Vortex Dynamics simulated using a lattice structure in C++, and I am currently working on a paper to publish my results. I also have industry experience working at Oracle as a Software Engineer within their Cloud Platform. Here I implemented features in Oracle's Serverless offering and learnt how software is built to be robust & efficient at scale. I gained experience working with technologies such as Kubernetes, Prometheus QL, Fluentd & Golang, and had exposure to the wider developer community by running hackathons and presenting at meetups. Studying abroad at the University of Washington gave me the opportunity to take further courses in supervised learning and complex systems, developing my interest in neural dynamics. This led me to pursue further research in the field of ML, particularly in Data-Intensive Science. During my PhD, I will be working with Nikos Konstantinidis to investigate pattern recognition and track reconstruction using Graph Neural Nets at ATLAS. I am looking forward to enriching my skillset, by the interdisciplinary nature of the CDT, which will enable me to pursue a future career within data science. 

---

Project description:  

In particle physics collider experiments, reconstructing particle trajectories (tracks) in the detector is computationally one of the most challenging parts of experimental data analysis. A typical LHC detector contains many thousands of sensors measuring particle positions, with a total number of sensor channels up to hundreds of millions. The track finding (aka pattern recognition) problem is to associate individual measurements (“hits”) into sequences representing particle trajectories. The scale of the problem is enormous, given that the number of hits can be up to 400000 per event; the number of tracks - several thousands. 

The current track finding algorithm adopted in the LHC experiments is based on the combinatorial track following from a subset of sensors into short track segments called seeds. As the seed number scales non-linearly with the number of hits, the corresponding, typically, close to cubical, CPU time increase creates huge and ever-increasing demand for computing power as the LHC will continue to increase the beam intensities over the next two decades. This motivates the research for novel methods in track finding, in particular those based on the machine learning (ML) techniques. The benefits of such an approach could lead to enormous savings in CPU needs over the next 20 years of life of the LHC. The aim of this project is to explore track finding methods utilizing a graph-based track model and graph neural networks (GNN), in order to predict compatible connections as well as iteratively extract track candidates. The main input features would involve track position measurements and cluster shape, which would be used to identify compatible hit-pair connections in the network. And the main techniques involved in pruning the graph network in order to extract track candidates would involve Gaussian mixture reduction via clustering, as well as using simplified Kalman filters as mechanisms for information propagation and track fitting. 

First year group project: ONS (Twitter)

Placement: 

---