Computer science has always had a sphere of influence that extends into many fields, and astronomy is just one of them. Recently, Ryan Hausen, a computer science graduate student, and Brant Robertson, an astronomy professor from the University of California, Santa Cruz, have teamed up to develop a machine learning model specializing in astronomy. Their work led to the creation of Morpheus, which is a deep-learning model designed to automate the classification of celestial objects. It is able to do this through learning from the prior classifications of real astronomers and applying that knowledge on new sets of data. Since the morphology of celestial objects can provide valuable information to astronomers, such as how galaxies are formed and evolve over time, they are motivated to increase the amount and rate of data observatories can gather. Methods of automation have become a necessity due to the sheer amount of data that modern–day telescopes and observatories are able to produce. Thus, astronomers turned to the field of computer science to drive the creation of Morpheus.
To understand how Morpheus is able to assist in processing imaging data from observatories and telescopes, knowing what machine learning is beforehand is necessary. In layman terms, they are algorithms that find patterns in data. It may not seem obvious, but this form of computer science has been subtly integrated into many aspects of daily life, including what posts show up on Instagram feeds and Netflix show recommendations. How it works is that they gather data of what sort of content the user frequently interacts with, and predicts the user’s future choices based on said data. When classified by its learning archetype (supervised, unsupervised, and reinforcement), Morpheus is supervised, due to the pre-existing classification data from astronomers and its programming dictating what sort of patterns it should look for. In Morpheus’ case, its programmers have taken it a step further by developing it as a deep learning network, hence its high degree of precision and power. In Morpheus’ case, prior classification of over 10,000 galaxies by astronomers from a 2015 study at the Hubble telescope is used to “train” the model to classify celestial objects accurately -- down to the last pixel.
Due to the amount of data that modern-day telescopes are able to acquire, it is unreasonable to expect astronomers to be able to classify each celestial object contained within said data. One such case of this issue is the Very Large Array (VLA) in New Mexico, which is currently running its own sky survey and producing imaging data at a rate of over 90 gigabytes per hour in the process. This calls for a more automated system of processing such data, and thus, is where computer science comes in. Prior to Morpheus, astronomers have used programming languages, including Python and C++, in order to create primitive data reduction systems. Though these systems undoubtedly helped in reducing the amount of data astronomers had to crunch through, the process of analyzing was still performed manually. The difference between simple reduction of prior systems and outright automated analysis of Morpheus is what sets them apart.
Though Morpheus is not the first image recognition algorithm that was attempted to be utilized in analyzing imaging data, Morpheus was the first one developed specifically for it. Its open-source nature means that the public is able to derive their own programs from it or even help develop it further. This means that the standard image format used as input by the algorithm is the original file generated at observatories. With this consistency comes the benefit of a degree of precision, down to the last pixel, as well as the ability to distinguish various shapes found in the imaging data; hence complex images are not an issue. In order to classify the various celestial objects found in imaging data, Morpheus generates more images with color–coding based on their morphology. However, the algorithm also recognizes that there can be a degree of inaccuracy within the classifications, and generates confidence levels for each accordingly.
Figure 1: Morpheus in action. Note the four separate color-coded images dedicated to identifying the various morphology of galaxies and the resulting classification image.
The sort of applications a machine learning model such as Morpheus, when used in this way, can have include the rapid mapping of the known universe. Thus, it can go a long way into answering several questions related to astronomy, how the universe began, and even when and where the universe ends. The majority of astronomers agree that constant expansion is the rule of the universe, which means that mapping and classifying astronomy objects accurately is crucial in order to measure and determine whether the law is true or false. One who looks at machine learning being applied in this way will be left to conclude that it is meant for individuals or organizations involved heavily in STEM. However, Morpheus and its impacts on astronomy only serve as an example of how the integration of computer science into daily life can ultimately be beneficial for humanity.
Bibliography
Altaweel, Mark. “Mapping the Universe.” GIS Lounge, 18 Feb. 2018, www.gislounge.com/mapping-the-universe/.
Hao, Karen. “What Is Machine Learning?” MIT Technology Review, MIT Technology Review, 17 Nov. 2018, www.technologyreview.com/2018/11/17/103781/what-is-machine-learning-we-drew-you-another-flowchart/.
Kent, Brian, and Joseph Masters. “Day and Night, We're Mapping the Sky (and Producing Terabytes of Data).” National Radio Astronomy Observatory, 30 Aug. 2018, public.nrao.edu/blogs/how-much-data/.
“Powerful New AI Technique Detects and Classifies Galaxies in Astronomy Image Data.” ScienceDaily, ScienceDaily, 12 May 2020, www.sciencedaily.com/releases/2020/05/200512151951.htm.
Storey-Fisher, Kate. “Morpheus, God of Dreams and Morphological Galaxy Classification.” Astrobites, 3 July 2019, astrobites.org/2019/07/03/morpheus-god-of-dreams-and-morphological-galaxy-classification/. Accessed 16 Aug. 2020.
Comentários