Artificial Intelligence (AI) systems have affected our world in many ways since their rise in the 1950s and has made a profound impact across a wide range of daily applications, making it one of the fastest-growing technologies globally. Its uses range from automating digital tasks and making predictions to enhancing efficiency and supporting smart language assistants.
Whether it be various businesses or architects, almost every other profession is leveraging AI to enhance productivity in their workflows. It is natural to question whether astronomers are utilizing AI to understand the universe better and, if so, what approaches they are taking. In fact, they have embraced the potential of Machine Learning (a subset of AI) since the 1980s, so much so that AI has become a standard part of the astronomer’s toolkit. This article highlights the eminent need for such systems in astronomical data analysis and dives deep into some recent applications where AI is employed.
The launch of the Hubble Space Telescope revolutionized the field of astronomy, yielding stunning imagery and essential data that has fundamentally altered our understanding of the universe. Today, driven by extraordinary advancements in AI, astronomy is experiencing ongoing evolution, uncovering significant insights that may elude human observation. Methods like Machine learning and neural networks have enabled classification, regression, forecasting, and discovery, leading to new knowledge and new insights.
The necessity of AI automation
A significant aspect of astronomy revolves around managing big data, where the term ‘big’ refers to Petabytes (1000 terabytes) and even Exabytes (1000 petabytes) of data collected from sky surveys like SDSS, Gaia, TESS and more. For instance, Gaia, a survey mission to map the Milky Way galaxy, collects approximately 50 terabytes of data each day. With the advancement of highly capable computer processing powered by AI, astronomers now possess the ability to analyze such massive volumes of data efficiently, significantly reducing the workload of scientists.
According to Brant Robertson, professor of astronomy at UC Santa Cruz, “There are some things we simply cannot do as humans, so we have to find ways to use computers to deal with the huge amount of data that will be coming in over the next few years from large astronomical survey projects.”
Even if all of humanity were to dedicate themselves to analyzing the vast amount of astronomical data, it would take an inconceivable long period to deduce meaningful conclusions. However, with the assistance of AI models, simultaneous processing and faster discovery of valuable information are possible, ultimately leading to increased efficiency and much shorter turnaround times. In addition, intelligent machines also improve accuracy and precision, where they can perform repetitive tasks with minimal to no errors.
The Emergence of AI in Astronomy
The utilization of AI techniques has evolved significantly over the years. A paper was published in 2020 titled; “Surveying the Reach and Maturity of Machine Learning and AI in Astronomy“, which discussed valuable insights into the historical progression of AI in this domain. Since the 1980s, principal component analysis (PCA) and decision trees (DT) have been employed for tasks such as morphological classification of galaxies and redshift estimation.
As the field advanced, artificial neural networks (ANNs) emerged as a widely used tool for galaxy classification and detection of gamma-ray bursts (GRBs) during the early stages of their implementation. The application of ANNs has since expanded to encompass diverse areas, including pulsar detection, asteroid composition analysis, and the identification of gravitationally lensed quasars.
Today, astronomers use a plethora of techniques that have resulted in exciting approaches involving the discovery of exoplanets, forecasting solar activity, classification of gravitational wave signals and even reconstruction of an image of a black hole.
I will explore three pivotal applications where the integration of AI plays a crucial role in solving complex problems, in turn shaping our understanding of the cosmos:
AI-Driven Morphology Classification of Galaxies
The classification of galaxies, whether they are elliptical, spiral, or irregular, enables us to gain insights into their overall structure and shape. This understanding is instrumental in estimating their composition and evolutionary trajectory, making it a fundamental objective in modern cosmology.
The advent of extensive synoptic sky surveys has led to an overwhelming volume of data that surpasses human capacity for scrutiny based on morphology alone. Since the 2000s, machine learning (ML) has appeared as the predominant solution to tackle this challenge and has effectively taken over the task of classifying galaxies. The classification of large astronomical databases of galaxies empowers astronomers to test theories and draw conclusions that reveal the underlying physical processes driving star formation and galaxy evolution.
The Deep Learning era brought forth Artificial Neural Networks (ANNs) that have accelerated the efficiency of classification and regression tasks by many folds. ANNs are computational models inspired by the human brain’s neural networks, capable of learning patterns and making predictions from large datasets. The input layer receives galaxy data, which is processed through hidden layers that perform complex computations. The output layer then generates classifications based on learned patterns. Each galaxy in the dataset is represented by a set of input features, such as photometric measurements or morphological properties derived from images.
While the vast volume of data can introduce model biasing, citizen scientists worldwide have collaborated through initiatives like Galaxy Zoo and Galaxy Cruise, playing a crucial role in validating the model results. This collective effort has effectively improved the accuracy of neural networks in classifying galaxies. Under the National Astronomical Observatory of Japan (NAOJ) project led by
Dr Ken-ichi Tadaki, ANNs have achieved an impressive accuracy level of 97.5%, where they identified spirals in about 80,000 galaxies. Thus, confirming the potential of AI systems in identifying the morphology of galaxies.
Reconstructing Black Hole images using Machine Learning
If you ask me what this century’s most remarkable scientific achievement is thus far, I would say that the black hole image revealed in 2019 would undoubtedly claim the top spot on the list. We get to see what a real Supermassive Black hole in Messier 87 looks like if we were there to see it.
Behind all the awe lies the immense dedication of the Event Horizon Telescope team, who invested two years in observing, processing, and eventually unveiling the black hole image to the public. Recently, the same data underwent a significant enhancement with Machine Learning, where we got a crisper, more detailed view of the light around the M-87 black hole. But then again, what was the need to use ML in the first place if we already got that incredible image back in 2019?
The Event Horizon Telescope is a network of eight radio telescopes in different areas of the globe, aiming to link them into a single array so that we can get an Earth-sized telescope. However, data gaps arise due to the irregular spacing between them, just like missing pieces in a jigsaw puzzle.
At first, scientists tried to blindly reconstruct the absent data from computer simulations and theoretical predictions. The image came up with model independence, which means that they did not assume they knew anything about what the final image should look like or had any idea of what shape it takes.
Without any presumed predictions, the team still managed to get a clear shape of a ring of light as Einstein’s theory of general relativity predicted. The appearance of a ring is attributed to the hot material orbiting the black hole in a large, flattened disc that becomes distorted and bends due to the black hole’s gravitational pull. As a result, this ring shape is observable from almost any viewing angle.
Now that we are pretty certain about what the image of a black hole should look like, scientists have developed a new technique called PRIMO (Principal-Component Interferometric Modeling) which uses sparse coding to find gaps in the input data. This algorithm builds on the initial data of EHT and more precisely fills in the missing gaps hence, achieving more resolution.
The newly reconstructed image is consistent with the theoretical expectations and shows a narrower ring with a more prominent symmetry. The greater the detail in an image, the more accurately we can understand the properties, such as the ring’s mass, diameter, and thickness.
Project lead author Lia Medeiros of the Institute for Advanced Study highlighted in her paper, “Since we cannot study black holes up-close, the detail of an image plays a critical role in our ability to understand its behaviour. The width of the ring in the image is now smaller by about a factor of two, which will be a powerful constraint for our theoretical models and tests of gravity.”
Techniques like PRIMO can also have applications beyond black holes. As Medeiros stated: “We are using physics to fill in regions of missing data in a way that has never been done before by using machine learning. This could have important implications for interferometry, which plays a role in fields from exo-planets to medicine.”
You can find more detail about the mentioned method in their paper published in The Astrophysical Journal letters.
AI’s Role in Detecting Water on Exoplanets
The study of extra-solar planets is one of the most fascinating and attractive fields of research in astronomy. As humans, our innate curiosity drives us to seek answers about the existence of life elsewhere in the universe. The exploration begins with the question of detecting water in exoplanets and other terrestrial bodies that might indicate the formation of life.
Astronomers have come across many techniques, most prominently spectroscopy, where the signatures of molecules in a celestial body can be detected. However, the time-intensive nature of spectroscopy creates a huddle for short observations. Therefore, there is a need for a simpler yet much more efficient method where the initial characterization of potential targets is separated before conducting detailed spectroscopic analysis at a later stage. This specific problem is being addressed by utilizing AI.
In a recent study, astrophysicists Dang Pham and Lisa Kaltenegger have used XGBoost, a gradient-boosting technique to characterize the existence of water in Earth-like terrestrial exoplanets in three forms; seawater, water clouds and snow. The algorithm is trained using the data of reflected broadband photometry, in which the intensity flux in specific wavelengths is measured from the reflected light of an exoplanet. The model shows promising results and achieves >90% accuracy for snow and cloud detection and up to 70% accuracy for liquid water.
In this way, a larger number of planets within the habitable zone having water signatures can be screened so that large projects like JWST can pinpoint and analyze extensively only the most favourable targets. According to Dr Pham: “By ‘following the water’, astronomers will be able to dedicate more of the observatory’s valuable survey time to exoplanets that are more likely to provide significant returns.”
Their recent publication of their findings can be found in the Monthly Notices of the Royal Astronomical Society.
(Ref: Pham, D., & Kaltenegger, L. (2022). Follow the water: finding water, snow, and clouds on terrestrial exoplanets with photometry and machine learning. Monthly Notices of the Royal Astronomical Society: Letters, 513(1), L72-L77)
Through ongoing research and advancement, AI continues to shape the future of astronomical exploration, enabling scientists to delve deeper into the vast expanse of the universe. Deep learning models like Convolution neural networks are revamping observational data in innovative ways, enabling discoveries even with data collected from older surveys.
We can only imagine what groundbreaking discoveries AI will bring when it is coupled with the powerful potential of the James Webb Space Telescope and upcoming projects like the Nancy Grace Roman Telescope. These visionary projects open doors to a realm of revolutionary discoveries, while the ever-expanding volume of astronomical data can now be harnessed to its fullest potential, thanks to the innovative advancements brought forth by the age of AI.
- Djorgovski, S. G., Mahabal, A. A., Graham, M. J., Polsterer, K., & Krone-Martins, A. (2022). Applications of AI in Astronomy. arXiv preprint arXiv:2212.01493.
- Fluke, C. J., & Jacobs, C. (2020). Surveying the reach and maturity of machine learning and artificial intelligence in astronomy. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 10(2), e1349.
- Impey, C. (2023, May 23). How AI is helping astronomers. EarthSky. Retrieved from https://earthsky.org/space/artifical-intelligence-ai-is-helping-astronomers-make-new-discoveries/https://www.astronomy.com/science/how-artificial-intelligence-is-changing-astronomy/
- Pham, D., & Kaltenegger, L. (2022). Follow the water: finding water, snow, and clouds on terrestrial exoplanets with photometry and machine learning. Monthly Notices of the Royal Astronomical Society: Letters, 513(1), L72-L77)
- Medeiros, L., Psaltis, D., Lauer, T. R., & Özel, F. (2023). The Image of the M87 Black Hole Reconstructed with PRIMO. The Astrophysical Journal Letters.
- NAOJ. (2020, August 11). Classifying Galaxies with Artificial Intelligence. Retrieved from https://www.nao.ac.jp/en/news/science/2020/20200811-subaru.html
Aly Muhammad Gajani holds a Master’s degree in Space Science and Technology specializing in Astrophysics together with GIS applications. His research focus is in the fields of galaxy evolution, astrophysical cosmology, exoplanet detection, and computational astronomy.