Need Deep Learning? There’s a Cloud for That.

February 29, 2016, 2:00 PM UTC
Image courtesy of Nervana.

After two years of effort and more than $24 million in venture funding Nervana Systems has opened up its deep learning cloud so any business can build computer models that can learn. Already customers are using it to help create artificially intelligent robots for farming, finding likely oil deposits using photos, and spotting fraudulent trading activity.

The startup, which is also building a specialized chip for deep learning, was founded by Naveen Rao, the former head of Qualcomm’s (QCOM) artificial intelligence chip efforts and two others who also left the mobile chip company to start Nervana. Their goal was to build a new type of processor modeled on the abilities of the human brain. But first they started with a cloud offering and everyday graphics processors to make money and test out software.

Today Nervana’s cloud is based on graphics processors purchased from Nvidia, but the founders of Nervana hope to replace the underlying hardware by the end of 2016 with specialized chips of their own design. Until then, the founders have both re-engineered the firmware that the Nvidia chips use and built their own software framework so deep learning jobs run faster on their cloud.

For more on artificial intelligence, watch:

The software framework is called Neon, and it competes with others such as Torch and Caffe, as well as those offered by Google (TensorFlow) and Microsoft (CNTK). Rao maintains that his, however, is 10 times faster running on the Nvidia (NVDA) hardware, even faster than Nvidia’s own framework. This speed matters because training a neural network and then running it is something that can take weeks or months. So any jump in speed, even if it’s halving the time, is a significant savings.

Steve Jurvetson, Partner at DFJ, who sits on the board at Nervana, believes that the evolution of deep learning is necessary for businesses to even make sense of the data they are gathering. This is where Nervana can help in the near term. “The world is awash in big data,” he says. “Just think about satellite imagery that allows you to count every tree or every car in a parking lot and imagine that every day someone had to try to count every car or tree and this more data than is possible for a human to handle. And this is where deep learning comes in.”

Get Data Sheet, Fortune’s technology newsletter.

For example, one of Nervana’s customers, Paradigm, which makes oil exploration software, has applied the same sort of computer vision training that helped Google’s deep learning networks learn how to identify a cat, and applied it to looking for oil. The company took thousands of seismic images and trained the neural network to look for certain types of faults where oil tends to pool and concentrate. Once trained, they ran their images through the Nervana cloud and got results back on where it might be worthwhile to drill. Such a painstaking search could take geophysicists years.

That’s today’s efforts. Rao, however, is thinking ahead. Like many others, he recognizes that the pursuit of artificial intelligence requires new chip architectures. So Nervana’s long-term effort is building a new type of semiconductor that takes some of its design elements from the human brain.

So far efforts, such as IBM’s, to build a literal brain in silicon have been limited by the material properties of silicon, the building block of today’s semiconductors. In the human brain information is carried to neurons by synapses and there are thousands of these per neuron. Replicating the information delivery structure in hardware requires too much wiring. So chipmakers are turning to software to build what’s known on chips as the I/O.

By taking a page from the software that runs the giant routers controlling the Internet, Nervana is trying to model statistically complicated information delivery virtually using software instead of physically using wires. The other important element for a deep learning chip is the ability to take a job and run it in parallel, something graphics processors are great at. That’s why so far Nvidia has an edge when it comes to deep learning, but, as many other startups point out, it doesn’t have the I/O.

Combining the two is what several researchers, startups, and big companies like IBM are trying to do so they can build the hardware to take AI more places. Meanwhile, on the cloud front, Nervana will compete with, which is also building a chip and offering a deep learning cloud; IBM’s Watson; MetaMind, which does image recognition; and Skymind, which offers deep learning for apps, and more.

Nervana Systems is backed by DFJ, DCVC, Allen & Company, AME Cloud Ventures, Andy Rubin’s Playground Global, CME Group, Fuel Capital, Lux Capital, and Omidyar Network. Customers so far include the Chicago Mercantile Exchange, farming robot creator Blue River Technology, and Paradigm.

Read More

Artificial IntelligenceCryptocurrencyMetaverseCybersecurityTech Forward