Topics: Coronavirus | AI | Data Science | Deep Learning | Machine Learning | Python | R | Statistics

KDnuggets Home » News » 2020 » Sep » Tutorials, Overviews » How AI is Driving Innovation in Astronomy ( 20:n37 )

How AI is Driving Innovation in Astronomy

In this blog, we look at a disruptive AI program - Morpheus - developed by Researchers at UC Santa Cruz that can analyze astronomical image data and classify galaxies and stars with surgical precision. If you're reading this with "starry" eyes, we bet we've got you hooked.

By Laduram Vishnoi, Founder at Acquire


Let's face it. If there's one buzzword that is taking several industries and professions by storm, it is Artificial Intelligence. But is it still a buzzword falling to deaf ears or has it gained wide-spread acceptance and momentum?

Data by PwC pegs the global impact of Artificial Intelligence at $15.7 trillion by 2030. On the other hand, Accenture claims that "Artificial Intelligence could double the rate of economic growth in developed countries by 2035." Needless to say, our money us on the latter.

Over the years, the term - "Artificial Intelligence" - or AI as it is commonly known - has been associated with a lot of things. And rightly so.

Siri, Alexa, robots, coding, Banking, E-Commerce, even immortality - and these are just a few. Clearly, the examples span the length and breadth of human imagination. However, there's one area that's relatively unexplored but equally exciting: AI in Astronomy. 

Consider these examples for a moment: 

In Japan, scientists are developing an Artificial Intelligence tool to predict the structure of the universe. Yes, you read that right.

Elsewhere, scientists are using 'smart' AI-powered telescopes to classify objects in space, with the ultimate aim of leveraging said telescopes to write and test hypotheses for physicists.

NASA's James Webb Space Telescope will soon be able to give users access to galaxies that were formed a couple of hundred million years after the Big Bang.

For the first time, a group of astronomers used artificial intelligence in a galaxy merger study to confirm that galaxy mergers were the driving force behind starbursts.  

Reading these real-life applications, the mind truly boggles. Furthermore, one thing becomes crystal-clear: 

More and more astronomers are using AI as a powerful discovery tool to offer rich and complex data, classify galaxies, sift through data for signals, find pulsar stars, identify unusual exoplanets, among other things. In short, there’s a whole new world of unexplored and infinite applications that are being experimented, thus giving rise to a slew of AI-powered astronomy tools better termed as"Artificial Astronomers." Give it some time. This phrase will catch on.

As you might've guessed by now, in this blog, we look at a disruptive AI program - Morpheus - developed by Researchers at UC Santa Cruz that can analyze astronomical image data and classify galaxies and stars with surgical precision. If you're reading this with "starry" eyes, we bet we've got you hooked.


AI in Astronomy: A New (Space)World Order



“One of the two main branches of science that extensively uses artificial intelligence is astronomy." - Somak Raychaudhury, a renowned astrophysicist and director of the Inter-University Centre for Astronomy

The Emergence of Morpheus:  The Origins of this (Inter)Stellar Computer Program 

Before we jump into the details, it makes sense to understand why there was a need to automate astronomy-related tasks in the first place. An astronomer by profession, Carlo Enrico Petrillo talks about the challenges of sifting through terabytes and terabytes of data. He says: 

“Looking at images of galaxies is the most romantic part of our job. The problem is staying focused.”

In the same vein, Brant Robertson, one of the developers of Morpheus and a professor of astronomy and astrophysics at UC Santa Cruz, explains:

"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."

It is this core idea that birthed project Morpheus which took around two years to come to fruition. Imagine if human astronomers were tasked with classifying space objects, they'd take eons (pun intended). But with AI software like Morpheus, it is possible to 'snap' objects with great accuracy and gather critical data on the evolution of galaxies with less than one second of thought. Or to put it in layman terms, you can literally understand how deep the universe' rabbit hole goes and view objects that came into existence a long time ago in a galaxy far, far away. Come on. You knew a Star Wars reference was coming. 

In terms of the effort and research that has gone into this project, you'll be surprised to know that the programmers used 10,000 galaxy images taken by NASA’s Hubble Space Telescope as ammunition (for lack of a better term) to better train the deep learning algorithm and the system by extension. Plus, large-scale surveys such as the Legacy Survey of Space and Time (LSST) will be used in conjunction with this program to understand the formation and evolution of galaxies. Just to give you a sense of what "The LSST can achieve, scientists claim that it will be able to take more than 800 panoramic images each night with a 3.2-billion-pixel camera, recording the entire visible sky twice each week." Interestingly, this CCD imaging camera will be able to produce 10 terabytes of data a night. Good luck finding astronomers who voluntarily want to sift through that data. This is where AI comes into play.

How the "Deep-Learning" Framework is Driving Morpheus 


"At the beginning of 2020, the digital universe was estimated to consist of 44 zettabytes of data. By 2025, approximately 463 exabytes would be created every 24 hours worldwide."

First things first, what do we mean by "Deep Learning?"

In simplistic terms, "Deep learning is a subset of machine learning where artificial neural networks, algorithms inspired by the human brain, learn from large amounts of data." If we were to draw parallels, you could think of deep learning as machines which 'learn' by repeating the same task as humans learn from experience. More often than not, small iterations are made on an ongoing basis to enhance outcomes each time a software uses deep learning algorithms.

Similarly, this program leverages deep learning and applies computer vision algorithms to classify objects based on the raw data streaming out of telescopes. Furthermore, it enables pixel-by-pixel classifications and brings to life a semantic segmentation of spatial objects - irrespective of their shape, whether they're disk, spheroidal, or irregular sized. Why is this helpful? Well, history indicates that morphologies of galaxies allows astronomers to understand how galaxies were formed and how they evolve over time.

Key takeaway: In a nutshell, using Morpheus, scientists can extract handy applications such as speech and image recognition to track galaxies, pixel-by-pixel.

Actual Processing by Morpheus in the Astronomical World: How this Well-Defined Technology Works

"With other models, you have to know something is there and feed the model an image, and it classifies the entire galaxy at once. Morpheus discovers the galaxies for you and does it pixel by pixel, so it can handle very complicated images, where you might have a spheroidal right next to a disk. For a disk with a central bulge, it classifies the bulge separately. So it’s very powerful.” - Brant Robertson

Here's a step-by-step guide of how Morpheus works:

Step 1: The program runs on UCSC's lux supercomputer and processes an image of a particular area in the sky.

Step 2: Based on this, a new set of images of that particular section are generated which are color-coded based on the object's morphology.

Step 3: The images clearly help identify stars and diverse types of galaxies. 

The final output: You get a pixel-by-pixel singular analysis for copious amounts of data, which is definitely a way approach to astronomical data analysis that anyone's tried in the recent past.

Advantages of Morpheus: A 360-Degree Approach

"When expert astronomers agree on the galaxy classification, Morpheus is 82 to 98 percent accurate depending on the class of object." - Brant Robertson

  • Offers Infinite Scope for Deeper Learning for Aspiring Astrophysicists: In a series of firsts, professors Brant Robertson and Ryan Hausen are releasing the Morpheus code publicly as well as offering online demonstrations. Plus, according to their research paper, tutorials for using the Morpheus deep learning framework have been created and publicly released as Jupyter notebooks. Here's an interactive visualization of the model.

An Image of a Large Disk Galaxy as seen on the Hubble Space Telescope  - Image Source



In Granular Contrast, the Morpheus Morphological Classification for the Same Region - Image Source


  • Provides granular detection and morphological classification of astronomical objects - which is virtually unheard of and humanly impossible. In fact, the model has the capability to recover over 98% of the galaxies that were present in the survey data that was used to train the model.
  • Allows for a powerful pixel-level classification by self-discovering galaxies and handling complicated images without human interference.
  • Offers the opportunity to learn about galaxy revolution in its entirety, without hindrances such as human bias or errors. Whether it pertains to understanding how galaxies evolved over time or even get a sense of where we're headed, this AI-empowered program is our best chance to understand the formation of galaxies in the purest form.
  • Eliminates chances of false-positive identifications of sources, which is a common phenomenon on the field of Astronomy.
  • Offers ease of use by supporting images in the commonly-used digital format for astronomical data: Flexible Image Transport System (FITS). You can say goodbye to converting telescopic images and data, and enjoy a friction-free experience.

Did You Know? Even with an ancient computer processor, AI-enabled gravitational lenses are capable of examining 21,789 images in just 20 minutes!


The Writing is on the Wall, /in Space, & Everywhere in Between

According to a Press Release by NASA, "The newly-discovered Kepler-90i – a sizzling hot, rocky planet that orbits its star once every 14.4 days – was found using machine learning from Google."

Clearly, AI's application in Astrophysics is offering 'astronomical' returns (pun intended) and redefining innovation in the world of Astro-Science, while helping uncover some of the biggest mysteries that lie in the depths of the universe. Stressing on the useful and collective culmination of AI and Astrophysics, Brant Robertson says: "Astronomy is on the cusp of a new data revolution", and we couldn't have summarized it better.

With Astronomers turning to AI to discover galaxies, they need not burn the midnight oil straining their eyes (and minds) to detect, classify, and decode spatial objects or hunt down new planets. In the 21st Century, we have AI-enabled super-telescopes that have their work cut out for them and no one's complaining. And, star-gazers, of course, are having a field day, rejoicing at the possibilities of using AI instruments to rediscover worlds beyond their already vivid imagination. Wonder what Elon Musk might have to say.
Bio: Laduram Vishnoi (@laduramvishnoi) is CEO and Founder at Acquire. He loves to share his research and development on Artificial intelligence, machine learning, neural network and deep learning. Read more at his Medium.


Sign Up

By subscribing you accept KDnuggets Privacy Policy