Hello, I am Brandon! I am an undergraduate at Penn State University Park where I am studying for my B.S. in Mathematics, and an intern at Argonne National Laboratory working on low-rank approximation problems.
Projects
I interned on and off at Brookhaven National Laboratory (BNL) from November 2020 until August 2022. During that time, I worked under Dr. Anže Slosar and Dr. Brett Viren on three seperate projects.
FRB Detection
I spent several months working on prototype algorithms for the real-time detection of fast radio bursts (FRB). An implementation of the algorithm ran briefly on Brookhaven's BMX telescope before getting replaced by a new and improved implementation.
WireCell Toolkit GPU Accelerated FFT Support
The second project I worked on was with Dr. Brett Viren.
During the fall of 2021, I spent several weeks writting a
package for
WireCell Toolkit
(WCT) which provided support for GPU-accelerated Discrete
Fourier Transforms using NVIDIA’s
cuFFT
library. This package has since been merged with WCT and
plays an important part in the collaboration's data
analysis.
DUNE Time-Slice Deghosting
The Deep Underground Neutrino Experiment (DUNE) aims to study the properties of neutrinos and relies heavely on several Liquid Argon Time Projection Chambers (LArTPC) to do so.
An LArTPC can be though of as large, parallel-plate capacitor filled with liquid argon. In the capacitor lie three planes of parallel wires, each rotated at a different angle.
When a neutrino enters the detector, it leave a trail of negatively charged particles. Such particles are pushed by the TPC's magnetic field towards the wire-planes. As these particles near a wire, they induce an electric current in that wire. This process is depicted in the figure below.
The detector “sess” a projection of the charged particles' positions. The process of reconstructing the electrically charged particles' possitions from the wire-plane waveform is called tomographic reconstruction. A naive tomographic reconstruction would lend many “possible” positions of neutrinos.
When I interned, there were two aproaches to resolve this issue. The first was a series of ad-hoc, combersome heuristics developed over an extended period of time. The second aproach was to use neural networks which were faster, but needed to be trained on inherently biased simulated data.
I was tasked with researching new algorithms for performing the tomographic reconstruction using the formalities of this paper. See the project's GitHub repo for more information.