SciTech

Algorithms help analyze astrophysical phenomenon

This is an artist’s representation of the completed Large Synoptic Survey Telescope, when it will be fully operational in 2016. Telescopes like these will be used to gather large amounts of observational data that will be analyzed for patterns by computer algorithms. (credit: Courtesy of Michael Mullen Design / LSST Corporation) This is an artist’s representation of the completed Large Synoptic Survey Telescope, when it will be fully operational in 2016. Telescopes like these will be used to gather large amounts of observational data that will be analyzed for patterns by computer algorithms. (credit: Courtesy of Michael Mullen Design / LSST Corporation)

Keeping track of the universe is no easy task; after all, the edges of the visible universe stretch nearly 93 billion light years across (a light year is about six trillion miles). Even so, dauntless researchers — at Carnegie Mellon University, Johns Hopkins University, and the University of Washington — are striving to do just that.

The U.S. Department of Energy has supplied a $1.6 million grant over a period of three years to fund the research of an automated method to detect astrophysical phenomena. The initiative is a collaborative effort between the three universities and is headed by Jeff Schneider, a research professor in Carnegie Mellon’s School of Computer Science. According to Schneider, limitations currently exist in the algorithms scientists use to analyze data gathered from powerful telescopes. “Current survey telescopes essentially scan the skies, log every image and object they see, and store them at data centers accessible to scientists around the world,” Schneider said.

Researchers have to prepare for the future, when even more powerful telescopes will be used that will gather greater amounts of data per day. One such telescope being built is the Large Synoptic Survey Telescope in Hawaii, which will be fully operational in 2016. Another is PAN-STARRS, also in Hawaii, which is already operational but expected to expand.
However, storing all of this data would take up petabytes of memory space — one petabyte is equivalent to one million gigabytes. Sorting through this much data, using current methods, would be impossible.

As a solution, Schneider’s research is intended to implement several methods to have machines sort through the data. This would incorporate both data mining methods, which is the process of finding patterns within data, and machine learning methods, which would allow computers to learn and make decisions based on given data. Schneider hopes to have machines implement anomaly detection, which has them “learn” a set of data about typical objects, and compare them with data from new observations.

Atypical objects would be marked as potential new discoveries. In addition, he hopes to implement unique pattern detection, a more global form of anomaly detection. As Schneider explained, “[Unique pattern detection] methods follow the same basic concept as anomaly detection, but they search for unique patterns over lots of objects rather than a single one. For example, maybe it would find that stars in galaxies that have the same size and shape as the Milky Way tend to also have higher proportions of metals in them.”

Currently, simulations exist that demonstrate events that take place in the universe. Schneider plans to create algorithms that will better match these simulations to observed activity in the universe. Further plans include using only data to create simulations to predict future events. “Here, the algorithm starts from scratch and has to figure out what the dynamic equations of a simulation should be in order to generate data like what we observe in the sky.” This research will hopefully bring about a better understanding of the universe, something so large, yet about which we know so little. Dark matter and dark energy, for example, may take up most of the observable universe, but our understanding is still unclear.

Astrophysics is not the only field that can benefit from this initiative. Schneider sees a larger picture for the future of his research: “All the other sciences, and especially the life sciences, are currently being deluged with the same avalanche of data that is simply too much for scientists to understand on their own. They need new discovery algorithms that will help them understand and find the knowledge in their data.” He notes that their work in science has become so complicated that, at times, it is necessary for computer algorithms, not scientists, to make discoveries. This research will, in effect, have machines analyze data and present findings to researchers. Schneider predicts that the future will hold even more complexity in terms of our understanding of the universe and the machines we use for research. “The details of the scientific models themselves may be too large and complex to be understood by individual scientists. This will call for even more sophisticated algorithms that are able to interact with scientists.”

There are countless discoveries waiting in the vastness of the universe. Schneider’s outlook for the future of astrophysics is hopeful: “We may find new kinds of objects that have never been observed before. We may find interactions that were previously unknown, such as the differing types of stars or galaxies that exist in very old or very young parts of the universe.”