Five years ago, Stanford Ph.D. student Richard Socher created the set of benchmarks that such companies as Google (GOOG), Microsoft (MSFT), and IBM (IBM) use to test their image-recognition systems. Now he has created a company that just might beat those standards.
MetaMind, the startup that Socher, after finishing his Ph.D. this summer, quietly co-founded with Sven Strohband, chief technology officer of Khosla Ventures, announced today that it has raised $8 million from Khosla (surprise) and Salesforce Chief Executive Marc Benioff. The pitch is so-called deep learning—the process of training networks of computers to resemble the human brain’s capacity to recognize and analyze visuals or text—as applied to businesses from medical imaging to fast food. In the latter case, Strohband says, “you take an image from your cellphone of the food you’re about to eat, and [the customer's computer system] automatically knows the food.” The company is giving away the technology on its website and charging interested users for help using and applying the tools.
MetaMind is jumping into a field crowded with big names investing orders of magnitude higher than $8 million. (Most, to be fair, keep their research to themselves.) And it’s not always easy to get lay customers to use the kinds of tools it’s selling even if they’re interested in the results, says Andrew Maas, a co-founder of Roam Analytics, which uses machine learning tools to build sales software. “The end user has to be smart enough that they’re going to be able to give you the data and set up the problem in a format that it is solvable by the techniques we have,” says Maas. “But at the same time, the user has to be not smart enough or not experienced enough to use the open source tools and run it themselves.”
Still, by at least one standard, MetaMind is already near the front of the pack. At this year’s ImageNet, the annual competition among deep-learning systems (using the standards Socher created), MetaMind’s setup correctly classified objects 92.4 percent of the time, second only to Google (GOOG).
(Bloomberg Businessweek)