Twitter Acquires Machine Learning Startup Whetlab

Next Story

Fitbay Doubles Down On Fashion Selfies

Twitter announced today that it has acquired Cambridge-based machine learning startup Whetlab in an effort to accelerate the company’s own in-house efforts on the matter. Deal terms were not immediately available, but Twitter will gain both access to Whetlab’s technology and small team following the acquisition, while Whetlab’s current product will be discontinued next month, the company notes.

Whetlab was developing A.I.-like technologies that would make machine learning easier for companies to implement. Its system had been designed to get a company’s internal systems off the ground automatically, and therefore, more quickly than before. This could potentially reduce the time it takes to train a new machine learning system from months to just days.

On its company website, Whetlab explains that their patent-pending system would offer an alternative to hiring experts to architect and tune a machine learning system in-house as is traditionally done, as it would do this work instead. The company even boldly claimed it did a better job than today’s human experts, noting that its tech had “outperformed the top machine learning researchers in configuring systems for the hardest cutting-edge problems.”

Screen Shot 2015-06-17 at 1.11.39 PM

The tech, which was developed by researchers at Harvard, Toronto, and Sherbrooke universities, was being put to use for things like object recognition, speech processing and even computational biology. It was running in a closed beta, according to a FAQ on the Whetlab website about the planned shutdown, which is taking place on July 15, 2015. The startup, a team of five, was only around a year old at the time of the acquisition.

All five team members, including Ryan Adams, Hugo Larochelle, Jasper Snoek, Kevin Swersky and Alex Wiltschko, are now joining Twitter where they’ll be tasked accelerating Twitter’s current machine learning efforts.

Whetlab, meanwhile, suggests that current beta testers download their data then migrate to other systems, like SpearmintSMAC, or HyperOpt, for example.

Machine learning technology, of course, could be put to a variety of uses at Twitter – it could learn to identify what users are posting about to enhance Twitter’s understanding of its user base; it could be used to improve spam detection systems; it could be used for better ad targeting; it could be used to make sense of Twitter’s massive data store; it could be used for recommendations and personalization; and much more.

At Twitter, we understand, the company is focused on using Whetlab’s technology for “organizing and surfacing relevant content.” However, the company isn’t yet sharing any information about how the technology, more specifically, will be put to use.

Whetlab’s own announcement on its plans now that it’s joining Twitter is fairly vague, noting only that: “Twitter is the platform for open communication on the internet and we believe that Whetlab’s technology can have a great impact by accelerating Twitter’s internal machine learning efforts.”