Researchers at the University of California, Santa Cruz (UCSC), led by astrophysicist Brant Robertson, are utilizing artificial intelligence and GPU-accelerated computing to process the immense data generated by modern telescopes, including the James Webb Space Telescope (JWST). This new approach addresses the challenge of analyzing an overwhelming volume of sensor readings, an amount far exceeding what human analysis or older CPU-based methods could handle. For instance, the JWST alone downlinks 57 gigabytes of imagery daily, while the upcoming Vera C. Rubin Observatory is expected to gather 20 terabytes nightly, and the Nancy Grace Roman Space Telescope will deliver 20,000 terabytes over its lifetime—a stark contrast to the Hubble Space Telescope's 1 to 2 gigabytes per day.
A key innovation is Morpheus, an AI system developed by Robertson and former graduate student Ryan Hausen. Morpheus, which is transitioning from convolutional neural networks to transformer architecture, processes large datasets to identify and classify galaxies. This system has already yielded surprising scientific results, including the discovery of rotating disk galaxies appearing much earlier in cosmic history than theoretical models predicted, a finding independently confirmed by other research groups. The team's AI pipeline, which runs on UCSC's Lux cluster (funded by a $1.6 million National Science Foundation grant) and an on-site Nvidia DGX Station, utilizes GPUs for every stage of the workflow, from data reduction and catalog generation to anomaly detection and simulation.
This high demand for GPU access by AI galaxy hunters contributes to a global GPU shortage. Robertson has been collaborating with Nvidia for 15 years to apply GPUs to astronomy, initially for supernova simulations and now for data analysis. The UCSC team is also developing generative AI models that can improve observations from ground-based telescopes by correcting for atmospheric distortion, a technique inspired by video game image reconstruction methods like Nvidia's DLSS. Future observatories, such as the proposed Habitable Worlds Observatory, will further intensify the need for scalable GPU-accelerated analysis, pushing universities to be more entrepreneurial in securing resources for cutting-edge AI and machine learning research.
