Abdulaziz Alkharjy

Artificial Intelligence Specialization · Cohort 2025

Machine Learning to Uncover the Oldest Light: U-Net Segmentation for Glitch Detection in Atacama Cosmology Telescope Data

The project develops and benchmarks advanced deep-learning modules that automatically clean and reconstruct the multi-terabyte timestreams recorded by the Simons Observatory and its precursor, the Atacama Cosmology Telescope. By suppressing instrumental noise and classifying foreground galaxies and stars, these models generate high-fidelity CMB maps while flagging potential astrophysical transients such as flaring stars and gamma-ray bursts. The work integrates convolutional, attention-based, and multi-modal techniques into the observatory’s production data-processing pipeline, laying the groundwork for scalable, real-time analysis of petabyte-scale cosmological data.

Mentor: Adam Hincks

Read the final report (PDF)

See it in the program report (PDF)