Now loading...
Brice Ménard, an astrophysicist from Johns Hopkins University and a researcher at Anthropic, has collaborated with Claude Science to create the first complete map of the sky in ultraviolet light. This groundbreaking map is poised to serve as a valuable educational resource, offering students detailed insights into the intricacies of the Milky Way as viewed through this specific wavelength.
The finished map encompasses the entire sky, with the galactic center prominently displayed. Notably, a significant portion of the image had previously been unobserved, and two-thirds of the new map was generated through a predictive method developed with Claude Science. Each pixel in the map is labeled to indicate whether it was measured or predicted, and uncertainty estimates are included, allowing for a comprehensive analysis of the data. For more details, you can view the map here.
In his astrophysics classes, Ménard often emphasizes how observations vary with different wavelengths, revealing distinct features of celestial objects. While visible light primarily captures stars, infrared highlights dust along the Milky Way, and radio waves expose hydrogen gas, the essential ingredient for star formation. In X-ray spectra, striking celestial events, such as supernovae and matter falling into black holes, come to light. These varying perspectives provide astronomers with a robust understanding of cosmic processes.
Previously, Ménard had limited options when discussing ultraviolet light, as available maps were incomplete and had significant gaps. The challenge with capturing UV light lies in its susceptibility to blockage by the ozone layer, necessitating space-based observations. Although a high number of space telescopes, including NASA’s GALEX mission from 2003 to 2013, have gathered data, large swathes of the sky still lacked comprehensive imaging.
The GALEX mission, which offered some two-thirds of the sky in around 38,000 observations, intentionally avoided bright star regions to prevent damage to its detectors, leaving vast blank spots. While statistical techniques exist to estimate these missing areas, the intricate work required often falls victim to researchers’ pressing priorities. However, Claude’s capabilities have made it easier to address such ongoing but lower-priority initiatives.
This past summer, Ménard embarked on the task of creating a complete UV map with the assistance of Claude Science. The project involved gathering all existing UV datasets, standardizing their metrics, and merging them to cover the unobserved sections of the sky. Claude used a team of AI agents to execute this mission. Their initial task was to scour for publicly available UV data, which involved processing a multitude of images taken over years under varying conditions.
After gathering the data, the agents ensured consistency within the surveys, correcting for any distortions caused by bright stars that might obscure fainter light. This uniformity was crucial before combining the datasets, which originated from different telescopes, each capturing the UV spectrum distinctively. High-level directives from Ménard guided Claude in executing the necessary tasks efficiently.
Filling in the gaps posed the most formidable challenge, with approximately one-third of the sky remaining unobserved in UV light. Ménard leveraged a technique known as inpainting, which allows a model to learn patterns in images and infer what missing parts might look like based on surrounding data. By correlating the partial UV observations with data collected at other wavelengths, including visible and infrared, the model could predict the characteristics of unobserved areas, complete with confidence metrics for each estimate.
To validate its accuracy, Claude was tasked with hiding sections of already mapped areas to simulate gaps, then asked to reconstruct these sections. After iterative refinements, Claude concluded that it could estimate hidden values with a mere 10% deviation from actual measurements, an imperceptible difference in visual terms.
In the final stages, Claude incorporated estimates of UV light from over 100 million stars based on visible light measurements provided by ESA’s Gaia satellite. Ménard was thrilled to present a complete UV sky map for the first time, showcasing a rich tapestry of dust clouds surrounding young stars, vast dust rings formed by stellar explosions, and faint filaments illuminated by starlight throughout the Milky Way. This comprehensive map stands to complement existing astronomical resources, enhancing educational opportunities for future generations of students.
The project’s success allowed Ménard to engage deeply in creating a significant scientific output without detracting from his other research commitments. Many scientists can relate to the struggle of advancing important projects that often linger due to other obligations.
