Machine learning startup Ayasdi raises $30.6M to map your data

Ayasdi, the machine learning startup that creates maps out of complex datasets, has raised a $30.6 million series B round of venture capital. New investor IVP led the round, with Citi Ventures and GE Ventures chipping in, as well as existing investors Khosla Ventures and Floodgate. This makes a total of nearly $41 million in funding for the red-hot Ayasdi, which emerged from stealth mode in January.

The company, founded by Gunnar Carlsson and Gurjeet Singh, uses a technique called topological data analysis to create the visually stunning maps that help set its product apart. A lot goes on under the covers in order to map out the data — hundreds of machine learning algorithms analyzing up to billions of data points — but the result is a map of the data that looks similar to a classic network graph. Only, instead of showing how data points are connected to each other, the clusters in the Ayasdi map signify similarity.

Aside from the sheer scale it can handle, the real benefit to the data analysts or data scientists who use the product, called Iris, is how it automates the process of discovering potentially valuable correlations. And it does so using many more algorithms that any one data scientist would likely use (or know how to use together) on a single dataset. With the results all right there in front of them, users can begin drilling down into the clusters to see whether there is indeed fire where the algorithms have discovered smoke.

“The amount of work they actually to do is reduced by a very large degree because they don’t have to write code anymore,” Co-founder and CEO Singh said.

One of Ayasdi's graph-like data maps
One of Ayasdi’s graph-like data maps

Ayasdi has attracted a lot of attention from large enterprise customers, including new new investors Citi and GE. The latter, Singh explained, is using Iris to find markers that can predict failure in jet engines. Oil and gas companies, such as Anadarko, are doing the same thing with drilling equipment, as well as analyzing vast amounts of “microseismic” data in order to figure out whether or not to drill in a new location.

Other Ayasdi customers include Merck, U.S. Food and Drug Administration, Centers for Disease Control and Prevention, the University of California San Francisco, Mount Sinai Hospital, Texas A&M University and Harvard Medical School. The company itself is now up to 50 people, Singh noted.

Asked how Ayasdi intends to evolve the platform, Singh said that a big data product always needs to increase the scale of data it can handle, and those like Ayasdi that target lines of business rather than the IT department can always get easier to use. “I don’t see machine learning as a field becoming dormant any time soon,” he said. “… We truly believe to make machine learning accessible, you have to make user experiences that are accessible to a very large number of people.”

Indeed, and Ayasdi isn’t alone it its approach to solving the world’s data problems, even if its approach is unique. We’ve covered different approaches to automatic pattern discovery by BeyondCore, as well as a handful of up-and-coming machine learning startups. An Atlanta-based company called Emcien is doing something at least similar to Ayasdi on the surface.

The goal is the same whatever the underlying technologies: take very complicated techniques and make them much easier to consume. If technology can mitigate the need for hordes of data scientists because it can empower business users, so be it. If technology just lets those hordes of data scientists operate a lot more efficiently and a lot more intelligently, that’s a win, too.

For an explanation on Ayasdi from Singh himself, check out this video of presentation from Structure: Data 2013 in March.


This post was updated at 5:36 a.m. to correct Gurjeet Singh’s title at Ayasdi.