Engineering x Data

By Grant Currin

“In the past, engineering was driven by a symbiotic relationship with the basic sciences,” says Garud Iyengar, professor of industrial engineering and operations research and director of Columbia’s Data Science Institute. “Engineering has always been data-driven, but as we developed better measurement technologies, we entered the big data regime.” 

In the hands of computer scientists and statisticians, big data became a raw input for making predictions that didn’t need to depend on understanding the underlying physics or chemistry of a system. This revolution began by tackling problems like labeling images of cats and, eventually, interpreting X-rays. Over time, researchers realized these technologies could be applied to a broad set of problems. “Suppose I want to build a new material. I know that this new material’s properties are going to depend on the molecules that comprise it,” Iyengar explains. “If I have data about tons and tons of molecules and how they behave, then I can start predicting what’s going to happen to the properties of a particular material.”

Over time, researchers have blended these approaches.

“Making predictions using data is great, but what if one could incorporate constraints based on limitations from physics and chemistry? By incorporating that knowledge, one can use the data much more efficiently and, in a sense, ‘regularize’ the prediction,” Iyengar says.

In a virtuous cycle, breakthroughs in measurement, data analysis, and engineering have collectively enabled an explosion in progress across every domain of engineering research. “I almost think of what is classically called data science as a step in the solution of an engineering problem,” Iyengar notes. “We are very fortunate at Columbia to have engineers who are thinking about the applied problems, plus really good statisticians, operations research people, and computer scientists who can come together to develop solutions that use data in the best, most responsible way possible.”