Understanding the universe’s vast history and predicting its future often relies on studying galaxies. Astronomers analyze the light galaxies emit – from visible light we see, to invisible ultraviolet and infrared – to learn about their composition, temperature, and age. But gathering these different “views” of the same galaxy is challenging, requiring expensive telescopes and lots of time. Now, a new AI-powered technique developed by a graduate student could allow researchers to quickly fill in missing pieces of this cosmic puzzle.
Contents
This innovative approach uses artificial intelligence to create high-quality predictions of what galaxies would look like in different types of light, even when observations are missing. It’s like having an AI artist who can paint a complete picture of a galaxy using just a few reference sketches. This could dramatically speed up research, make telescope time more efficient, and open new avenues for discovery in astronomy.
Seeing Galaxies in a New Light
Imagine looking at the world through different colored filters – a red filter might highlight some details, while a blue filter shows others. Galaxies are similar, but instead of colored filters, astronomers use instruments that capture light in different wavelengths, called photometric bands. Visible light is just one band; ultraviolet (UV) light reveals hot, young stars, while infrared (IR) light shows dust and cooler objects. Combining these views gives astronomers a complete picture of a galaxy’s life cycle and structure.
Collecting data across all these bands for countless galaxies is a monumental task. Existing ground and space-based telescopes provide snapshots, but often data is incomplete. This is where Youssef Zaazou, a graduate student at Memorial University, saw an opportunity to apply machine learning.
Illustration showing how astronomical objects appear different in ultraviolet, visible, and infrared light
Working in both Mathematics & Statistics and Computer Science, Zaazou developed a generative deep learning model. Generative AI is a type of artificial intelligence that can create new content, like images or text, based on patterns it learns from existing data. In this case, the AI learned how galaxies look in different wavelengths.
Teaching AI to See Like a Telescope
Zaazou’s technique, detailed in his paper “Mapping Galaxy Images Across Ultraviolet, Visible and Infrared Bands Using Generative Deep Learning” and published in the American Astronomical Society Journals, is the first of its kind. It trains the AI model to take an image of a galaxy in one or two wavelengths and generate realistic, high-quality images of what that same galaxy would look like in the other wavelengths.
“This project bridges my love for astronomy and my interest in machine learning,” says Zaazou. “I wanted to be an astronomer, but I realized it was somewhat impractical. This project is a good concession.”
Before this, there wasn’t an efficient way to predict what a galaxy would look like in a missing band without actually observing it. Astronomers had to allocate precious telescope time, which is always in high demand, to capture that specific data. Zaazou’s AI model offers a way to bypass some of these observational gaps.
“No one has attempted to transform galaxy photos across photometric bands using AI before,” Zaazou explained. “Generative AI remains a relatively underutilized tool in astronomy. We hope that this work will motivate the astronomical community to incorporate AI into their toolbox.”
The efficiency is remarkable. Zaazou noted that his model can generate about 2,000 images in roughly 10 minutes, and the model itself is quite small, only about 10 megabytes. This low computational cost means it can be used widely without needing supercomputers.
From Campus to the Cosmos
This research demonstrates the powerful potential of AI beyond just text or image generation; it can automate complex tasks in fundamental science like astronomical image translation. By providing missing views of galaxies quickly and cheaply, this AI can help astronomers make better use of their telescope time, focusing on observing completely new objects or gathering even deeper data, rather than filling in known gaps.
For Zaazou, this project was also deeply personal. He spent months analyzing galaxy images, so much so that he even dreamt about them.
Graduate student Youssef Zaazou, who developed an AI technique for mapping galaxy images across different wavelengths
He credits his supervisors, Drs. Alex Bihlo and Terrence Tricco, for their guidance and support, which was crucial in bringing this novel project to fruition. Their knowledge and proactive support helped him navigate the challenges of such interdisciplinary work.
“They’ve been incredible. Their knowledge, contributions and moral support made this work possible,” he said.
The Future is Bright (in Many Wavelengths)
Zaazou is optimistic about the impact of his work. He believes that demonstrating this successful application of generative AI in astronomy will encourage more researchers to explore how these powerful tools can aid in studying the cosmos.
“Hopefully, this sparks more investigation into how AI models can aid astronomers,” he stated. “They’re still underutilized right now.”
His work highlights how combining computer science with traditional scientific fields can lead to breakthroughs that accelerate our understanding of the universe. It’s a reminder that some of the most exciting discoveries happen at the intersection of different disciplines.
Illustration showing how astronomical objects appear different in ultraviolet, visible, and infrared light
As Zaazou looks to the future, he reflects on the privilege of using his technical skills for something as awe-inspiring as studying galaxies. It’s a connection between the complex world of AI and the vast wonders of the universe, making the impossible just a little bit closer. This new AI tool could soon be helping astronomers around the world piece together the universe’s grand story, one galaxy image at a time.
