UNIVERSITY PARK, Pa. — Combining astrophysical knowledge with machine learning techniques can address problems that can’t always be solved using domain expertise or computational methods alone, according to Eric Ford, a collaborative employer at Penn State’s Institute for Computational and Data Sciences (ICDS) and distinguished professor of astronomy and astrophysics.
Ford, who joined Penn State in 2013, works on a balanced combination of statistical, computational and machine learning techniques to detect low-mass, potentially Earth-like planets orbiting stars outside our solar system.
“I am with ICDS [astronomy and astrophysics] “Penn State faculty played a major role in some of the early discoveries of exoplanets and have built one of the premier exoplanet programs in the world. The combination of Penn State's ICDS, the Center for Exoplanets and Habitable Planets and the Center for Astrostatistics aligned very well with my research interests,” Ford said.
Ford said he had been interested in exploring planets in the solar system since he was a child, but it wasn't until he got to college that it all started to fall into place.
“I was still in college when the first exoplanet orbiting a sun-like star was discovered,” Ford says. “The field was just starting out, so I was really happy to be able to become almost as much of an expert as my professors. I really loved the vibrancy of the field, and that's what got me started on my path in exoplanet research.”
Ford's ongoing research projects with students and postdoctoral researchers are using multiple approaches to find Earth-like planets.
“We use spectrometers to look at the spectrum of a star and infer the velocity of the target star due to an orbiting planet,” Ford said. “We are working to improve on existing techniques to detect potentially Earth-like planets.”
According to Form, measuring the masses of these potentially Earth-like planets is essential to interpret future observations of exoplanet atmospheres and characterize their habitability.
“For many years, astronomers were limited to inevitable variations in a star's spectrum caused by variations in the amount of starlight collected and the instruments used to collect the data,” Ford added. “With a new generation of highly stabilized spectrometers, such as the HPF and NEID spectrometers built at Penn State, we are now sensitive to even smaller variations caused by the stars themselves.”
Ford noted that stars have specks, pulsations and convection patterns that also cause variations in a stellar's measured spectrum.
“Machine learning is used to distinguish which variations are due to planets orbiting the star and which are due to stellar processes,” Ford said. “Much of my work involves applying modern statistical methods to astronomy. My approach is to first ask, 'What would we want to do in an ideal world?' and then figure out how to jump through computational hurdles to answer the question using solid statistical foundations. Sometimes this means combining high-fidelity and low-fidelity models. Machine learning makes some computations more efficient, allowing us to devote more resources to other aspects of the problem where machine learning is less well suited.”
Because of the complexity of the research, Ford said he took multiple approaches and tried to stay true to the interdisciplinary research mission of ICDS and Penn State.
“One of the great things about interdisciplinary research is that researchers in fields like computer science and artificial intelligence are trying so many new things,” Ford says. “They have their own goals, which often aren't directly applicable to the physical sciences. But their work often contains valuable nuggets of insight that can be applied to benefit astronomy. The trick with interdisciplinary research is that you can't do all the calculations needed to try every idea. You have to learn a little bit about the many possible methods, select those, and spend more time maturing them. It's both frustrating and fun.”
Ford said the results of research and training future generations were equally important.
“Some of the students we mentor go on to become astronomers, while others apply their data science and computational skills to other fields like energy or environmental monitoring,” Ford says. “The scope for progress is to discover a wide variety of exoplanets, improve our understanding of their formation, and learn whether our solar system is one in a million or if there are others like it. As we learn more about other planetary systems and their formation, those insights will trickle back into information about how our solar system formed and what is special about ours.”
