Cosmologists are increasingly using advanced statistical methods and algorithms to tackle challenging questions about the universe. But analysis of the torrents of data from current and future observatories may also benefit from the use of innovative machine learning techniques.

To explore possible multidisciplinary opportunities, Argonne researchers organized a three-day workshop titled ​“Advanced Statistics meets Machine Learning III.”  

The workshop, held at Argonne on November 13–15, 2019, covered topics ranging from Bayesian inference and uncertainty quantification to unsupervised deep learning and methods for dealing with noisy and low-dimensional data. In addition to talks, the workshop featured hands-on sessions, with participants working on problem statements and data provided beforehand.

 ​“Throughout, our focus was on how statistics and machine learning together can help solve cosmological applications,” said Arindam Fadikar, a postdoctoral appointee in the Mathematics and Computer Science Division at Argonne and co-organizer of the workshop. ​“New statistical data analysis tools, coupled with recent developments in machine learning techniques, show promise in evaluating the enormous amount of data being accumulated by sky surveys.”

The workshop brought together participants from national laboratories, universities and industry, who presented their current approaches and discussed both the challenges and possible limitations of combining machine learning methods and computational statistics to interpret large cosmological datasets.   

“The sessions culminated in several fruitful discussions leading to new insights into inference problems in computational cosmology and fostering new perspective into existing statistical and machine learning techniques,” Fadikar said.