Select Country
Follow Us

This Astronomer’s Idea Just Opened A Universe Of New Opportunities For GE’s Digital Growth

Before GE acquired, the machine learning company saw the future written in the stars.

In 2008, Joshua Bloom, a professor of astronomy at the University of California, Berkeley, was struggling to make sense of tens of thousands of telescope images of the night sky — black pictures dusted with little white dots. The differences between the photos were too subtle for the human eye.

When Bloom shared his problem with colleagues from the faculty’s statistics and computer science departments, they pointed him to machine-learning software — sets of powerful algorithms that can crunch through mountains of data and spit out answers in seconds.

When Bloom unleashed the software on his millions of white dots, “he ended up building a more powerful telescope that was able to better understand some of the rarest phenomena in the universe,” says Jeff Erhardt, the chief executive of, referring to marvels such as white dwarfs and other strange stars.

It was a “eureka” moment for Bloom, who with Erhardt went on to found and is now the Berkeley-based company’s chief technical officer. uses those same machine-learning and artificial-intelligence techniques back on planet Earth. The company makes highly sophisticated software that helps the business and industrial worlds make sense of their own galaxies of data.

This picture is an artist's impression showing how the binary star system of Sirius A and its diminutive blue companion, Sirius B, might appear to an interstellar visitor. The large, bluish-white star Sirius A dominates the scene, while Sirius B is the small but very hot and blue white-dwarf star on the right. The two stars revolve around each other every 50 years. White dwarfs are the leftover remnants of stars similar to our Sun. The Sirius system, only 8.6 light-years from Earth, is the fifth closest stellar system known. Sirius B is faint because of its tiny size. Its diameter is only 7,500 miles (about 12 thousand kilometres), slightly smaller than the size of our Earth. The Sirius system is so close to Earth that most of the familiar constellations would have nearly the same appearance as in our own sky. In this rendition, we see in the background the three bright stars that make up the Summer Triangle: Altair, Deneb, and Vega. Altair is the white dot above Sirius A; Deneb is the

Above: Bloom first used his software to classify white dwarf stars like Sirius B (on the right) in the Sirius binary star system located in our cosmic backyard. Illustration credit: Getty Images. Top image: A planetary nebula created by a star going supernova. This SN 1006 supernova remnant is located about 7000 light years from Earth. Image credit: Getty Images

The software builds predictive models based on maps of past patterns. These models are continually updated as the computer receives fresh information. Put that kind of software to work on a huge industrial machine, and companies will discover efficiencies they did not even know were up for grabs, Erhardt says.

How big are the savings? GE Digital expects software and analytic tools, which now include, to bring in $15 billion in revenue by 2020, $1 billion of which will come from increased efficiencies.

Specifically, will beef up the machine-learning apps GE already has on its Predix platform for the Industrial Internet, including the digital twin, which builds a virtual copy of any piece of equipment.

Today, engineers can use the twins to model different scenarios and decide when a piece of equipment needs to be serviced and when it can stay in use.’s machine learning will allow some of those decisions to happen without human intervention. Just like it did when it was analyzing the stars,’s software will be able to find patterns that aren’t apparent to the human eye and make decisions and changes on the fly., which GE bought for an undisclosed sum, already works with companies such as Pinterest, Citrix and Volkswagen. At Pinterest, for example, analyzed thousands of help requests and found patterns that helped the social media company resolve customer problems much more quickly.

For example, inspectors at an oil and gas pipeline company can now take reams of data from acoustic pressure measurements and filter out unnecessary noise to focus on data that might signal the risk of a dangerous leak. This reduces the time to generate an inspection report from weeks to days, Erhardt says.

GE will be using’s software across its many divisions, including healthcare, where machine learning could drastically improve diagnoses. “Machines generate hundreds of thousands of images,” says Erhardt. “The technology could augment and automate the process used to look for anomalies in a tumor or a cancer.”

Subscribe to our GE Brief