Abby Tejera Rocha

Abby Tejera Rocha

Biography

Abby Tejera is a recent graduate with double majors in Physics and Computer Science from Oberlin College. There, she was a student representative of the Physics and Astronomy department and a member of her college’s Computer Science Majors Committee. She is a co-founder of Bridging Resources and Access to Nurture Community Through Holistic Engagement in STEM (BRANCHES), a committee that fosters community in STEM-related disciplines. She has conducted research on metallicities in different galaxies in Professor Scudder’s group. She also conducted research at the Laser Interferometer Gravitational-Wave Observatory (LIGO)under the mentorship of Dr. Gabriela González. Abby previously participated in MSRP in 2024, working on astronomy research under the mentorship of Dr. Anna-Christina Eilers. Now in 2025, she is back at MIT researching how to utilize AI for neutrino detection under the mentorship ofDr. Michael Williams and Dr. Jessie Micallef in the MIT Institute for Artificial Intelligence and Fundamental Interactions (IAIFI). Abby will be pursuing a PhD in Astrophysics and Astronomy at Johns Hopkins University and plans to continue her research as a member of LIGO. Outside of her research, Abby enjoys participating in community service initiatives and engaging in science dissemination initiatives.

Abstract

Multi-Detector Machine Learning Reconstruction for Enhanced Neutrino Physics in Liquid Argon and Solid Scintillator Detectors

Neutrinos are some of the most mysterious particles in the universe, tiny, nearly massless, and incredibly hard to detect. The Deep Underground Neutrino Experiment (DUNE) is a large-scale international collaboration aiming to better understand properties of neutrinos by sending a beam of neutrinos from Fermilab to detectors 800 miles away. This research explores how machine learning can make neutrino detection more accurate. A prototype for one of DUNE’s near detectors (situated near the neutrino beam source), which uses a liquid Argon target, has taken some preliminary data in 2024. Currently, the detector uses a Graph Neural Network (GNN) for particle reconstruction. Our work focuses on incorporating data from a nearby detector that uses a solid scintillator target into that same GNN to better match particle tracks between detectors, avoid overfitting, and correctly identify different particle types. Testing these improvements on the prototype is a crucial step before scaling up to DUNE’s full near detector. By using AI to improve how we reconstruct what happens during neutrino interactions, this work brings us closer to unlocking the secrets of these elusive particles—and, in turn, the universe itself.