Is Astronomy Data Science?
Machine learning in astronomy may sound like an oxymoron, but is it really? One of the most recent ‘sciences’ is machine learning, while astronomy is one of the oldest. In fact, astronomy evolved naturally as people discovered that studying the stars is not only fascinating but can also help them in their daily lives. Investigating the star cycles, for example, aided in the creation of calendars (such as the Maya and the Proto-Bulgarian calendar). Furthermore, it was critical for navigation and orientation. The use of mathematical, geometrical, and other scientific techniques to analyse observed data was a particularly significant early development. It all began with the Babylonians, who established the astronomical traditions that would later be followed by many other civilizations. Since that time, data analysis has played an important role in astronomy.
So, after millennia of refining data analysis techniques, you’d think that no dataset could pose a problem to astronomers anymore, right?
That’s not entirely correct. The main issue that astronomers are currently dealing with is… as strange as it may sound… technological advancements.
What the heck?! How can improved technology be a hindrance? It most emphatically can. Because what I mean by better technology is a larger field of view (FOV) of telescopes and higher detector resolution. All of these factors indicate that today’s telescopes collect far more data than previous generations of technology. That means astronomers will have to deal with data volumes they’ve never seen before.
How Did The Galaxy Zoo Project Come To Be?
Kevin Schawinski found himself in a similar situation in 2007.
As an astrophysicist at Oxford University, one of his responsibilities was to classify 900,000 images of galaxies collected by the Sloan Digital Sky Survey over a seven-year period. He had to examine each image and determine whether the galaxy was elliptical or spiral, and whether it was spinning. The task appears to be quite simple. However, the sheer volume of data made it nearly impossible. Why? Because, according to estimates, it would have taken one person 3-5 years to complete! Talk about a massive workload! So, after a week of working, Schawinski and his colleague Chris Lintott decided there had to be a better way.
That’s how Galaxy Zoo, a citizen science project, came to be. If you’re unfamiliar with the term, citizen science refers to the public’s participation in professional scientific research. Schawinski and Lintott’s basic plan was to distribute the images online and recruit volunteers to help label the galaxies. And this is possible because determining whether the galaxy is elliptic or spiral is a simple task.
They had hoped for 20,000 to 30,000 people to contribute at first.
Surprisingly, more than 150,000 people volunteered for the project, and the images were classified in about two years. Galaxy Zoo was a success, and other projects such as Galaxy Zoo Supernovae and Galaxy Zoo Hubble followed. To this day, there are several active projects.
Using thousands of volunteers to analyse data may appear to be a success, but it also demonstrates how much trouble we are currently in. In just two years, 150,000 people were able to classify (rather than perform complex analysis on) data collected from a single telescope! And now we’re building telescopes that are a hundred, even a thousand times more powerful. However, in a few years, volunteers will be insufficient to analyse the massive amounts of data we receive.
To put this into perspective, the rule of thumb in astronomy is that the amount of data we collect doubles every year. For example, the Hubble Space Telescope, which has been in operation since 1990, collects approximately 20GB of data per week. The Large Synoptic Survey Telescope (LSST), which is set to launch in early 2020, is expected to collect more than 30 terabytes of data per night.
However, this pales in comparison to the most ambitious project in astronomy, the Square Kilometre Array (SKA). SKA is an intergovernmental radio telescope that will be built in Australia and South Africa and is expected to be completed in 2024. It is expected to produce more than 1 exabyte per day with its 2000 radio dishes and 2 million low-frequency antennas. That’s more than the entire internet for an entire year, created in a single day!
Machine Learning In Astronomy, It Turns Out, Is Also A Thing. Why?
To begin with, machine learning processes data much faster than other techniques. However, it can also analyse that data for us without our assistance. This is critical because machine learning can understand things we don’t even know how to do yet and recognise unexpected patterns. It could, for example, distinguish between different types of galaxies before we even know they exist.
This leads us to the conclusion that machine learning is less biassed than humans and, as a result, more reliable in its analysis. For example, we may believe there are three types of galaxies out there, but a machine may see five distinct ones. And this will undoubtedly improve our rudimentary understanding of the universe.
Regardless of how intriguing these issues are, the true strength of machine learning is not limited to solving classification problems. In fact, it has much broader applications that can extend to previously unsolvable problems.
What Exactly Is Gravitational Lensing?
In 2017, a Stanford University research group demonstrated the effectiveness of machine learning algorithms by studying images of strong gravitational lensing with a neural network.
Gravitational lensing is an effect in which the strong gravitational field surrounding massive objects (such as a galaxy cluster) bends light and produces distorted images. It is one of Einstein’s major predictions from his General Theory of Relativity. That’s all well and good, but you might be wondering why studying this effect is useful.
What you must understand is that ordinary matter is not the only source of gravity. Scientists propose that the majority of the universe is made up of “invisible matter,” also known as dark matter. However, we cannot directly observe it (hence the name), and gravitational lensing is one method for “sensing” and quantifying its influence.
This type of analysis was previously a time-consuming process that involved comparing actual lens images with a large number of computer simulations of mathematical lensing models. For a single lens, this could take weeks or months. That is what I call an inefficient method.
However, using neural networks, the researchers were able to perform the same analysis in just a few seconds (and, in theory, on the microchip of a cell phone), which they demonstrated using real images from NASA’s Hubble Space Telescope. That’s quite impressive!
Overall, the ability to sift through large amounts of data and perform complex analyses quickly and fully automated could transform astrophysics in ways that are critical for future sky surveys. And these will delve deeper into the universe, yielding more data than ever before.
What Are The Current Applications Of Machine Learning?
Now that we know how powerful machine learning can be, it’s natural to wonder: Has machine learning in Astronomy been used for anything useful yet?
The answer is… sort of. The truth is that the use of machine learning in astronomy is a relatively new technique. Although astronomers have long used computational techniques to aid in their research, such as simulations, ML is a different beast.
Nonetheless, there are some examples of ML in action.
Let’s start with the simplest. Telescope images frequently contain “noise.” We define noise as any random fluctuations that are unrelated to the observations. Wind and the structure of the atmosphere, for example, can affect the image produced by a ground-based telescope because air gets in the way. That is why we send some telescopes into space: to remove the influence of the Earth’s atmosphere. But how do you cut through the noise created by these factors? Using a machine learning algorithm known as a Generative Adversarial Network (GAN).
GANs are made up of two parts: a neural network that attempts to generate objects and another (a “discriminator”) that attempts to determine whether the object is real or fake-generated. This is an extremely common and successful noise-reduction technique that is already dominating the self-driving car industry. It is critical in astronomy to have as clear an image as possible. That is why this technique is becoming more popular.
NASA Is Another Example Of Artificial Intelligence
This time, however, it has non-space applications. I’m referring to wildfire and flood detection. Using satellite images, NASA trained machines to recognise smoke from wildfires. What is the goal? Hundreds of small satellites will be launched, each with machine-learning algorithms embedded within sensors. With this capability, the sensors could detect wildfires and transmit data back to Earth in real time, providing firefighters and others with up-to-date information that could significantly improve firefighting efforts.
Is There Anything Else You Want?
Yes, NASA is conducting research on the critical application of machine learning in probe landings. Sending probes to land on asteroids, collect material, and return it to Earth is one method of space exploration. Currently, in order to select a suitable landing site, the probe must take pictures of the asteroid from every angle, send them back to Earth, and then scientists manually analyse the images and instruct the probe on what to do.
This intricate process is not only complex, but also somewhat limiting for a variety of reasons. To begin with, it is extremely taxing on the project’s participants. Second, keep in mind that these probes could be thousands of miles away from home. As a result, the signal carrying the commands may have to travel for minutes or even hours to reach it, making fine-tuning impossible. That is why NASA is attempting to cut the “informational umbilical cord,” allowing the probe to recognise the asteroid’s 3D structure and select a landing site on its own. And neural networks are the means to accomplish this.
What Challenges And Limitations Do Machine Learning In Astronomy Face?
Why has it taken so long for machine learning to be used if it is so powerful? One of the reasons is that in order to train a machine learning algorithm, a large amount of labelled and processed data is required. Until recently, there simply wasn’t enough data for a computer to study some of the more exotic astronomical events. It should also be noted that neural networks are a kind of black box in that we don’t fully understand how they work and make sense of things. As a result, scientists are understandably concerned about using tools without fully comprehending how they work.
While we at 365 Data Science are very excited about all ML developments, it is important to note that it has some limitations.
Many people assume that neural networks are much more accurate and have little to no bias. Though this is true in general, it is critical for researchers to understand that the input (or training data) they feed to the algorithm can have a negative impact on the output. The training set is used to teach AI. As a result, any biases in the initial data, whether intentional or unintentional, may persist in the algorithm.
For example, if we believe there are only three types of galaxies, a supervised learning algorithm will also believe there are only three types of galaxies. As a result, even if the computer does not introduce additional bias, it may end up reflecting our own. That is, we can train the computer to think in a biassed manner. As a result, ML may be unable to identify some revolutionary new model.
These are not game-changing factors. Nonetheless, scientists who use this tool must consider these factors.
So, What Is The Future Of Machine Learning?
The data we generate is increasingly shaping the world in which we live. As a result, it is critical that we incorporate data processing techniques (such as machine learning) into all aspects of science. The more researchers who use machine learning, the greater the demand for graduates with experience in it. Even today, machine learning is a hot topic, and it will only get hotter in the future. And it remains to be seen what milestones we will achieve with AI and ML, as well as how they will transform our lives.
