Artificial Intelligence and Machine Learning

You are here

From internet search engines to autonomous vehicles, artificial intelligence (A.I.)—once mostly the stuff of science fiction—is becoming an increasingly familiar and practical feature of everyday life. 

It’s also having a growing impact on science—and has become an important area of interest for the DOE Office of Science.

Two factors are driving interest in A.I. across the globe: the vast accumulation of data in our increasingly digitized society, and the growing capability and speed of today’s fastest computers for processing that data.

A.I., including machine learning (ML), is an ideal tool for deriving new insights from analysis of very large data sets. A.I. becomes more useful as the speed and computational power of today’s supercomputers grows. 

DOE currently owns the world’s most capable such machines. The newest of these—Summit at Oak Ridge National Laboratory—has an architecture especially well-suited for A.I. applications.

Science itself, meanwhile, is increasingly driven by “big data.” DOE Office of Science user facilities, such as particle accelerators and X-ray light sources at the DOE national laboratories generate mountains upon mountains of data. It is believed that A.I. is poised to be a powerful new tool for analyzing this data and deriving discoveries from it. Using basic machine learning, researchers are identifying patterns or designs that are difficult or impossible for humans to detect, at speeds that are hundreds to thousands of times faster than traditional data analysis techniques. 

The most interesting scientific applications of scientific machine learning are those, such as materials discovery or high-energy physics, where the answers are unknown beforehand or the results of an automated system are not easily verified. DOE is supporting development of new methods and algorithms that increase the reliability (for what type and quantity of data do we expect results), robustness (how might slightly different data change the results), and rigor (have the assumptions and underlying theories been defined and validated) of machine learning algorithm and methods to support their use in scientific research.

In the meantime, DOE is currently partnering with the National Institutes of Health to use DOE’s supercomputers for A.I. analysis of the government’s vast store of cancer patient data to improve treatments. DOE is also partnering with the Veterans Administration, using A.I. on Veterans’ health data to impact suicide prevention, prostate cancer, and cardiovascular disease.  Both efforts are expected not only to yield important health benefits, but also to help advance the design of A.I. algorithms and applications.