Twitter enters deep learning race with MadBits

As web giants fight to define next generation user experience and expand big data potential, Twitter buys imaging AI start-up

By Caroline Gabriel

One of the hottest areas for M&A this year is deep learning, as the web giants build up their engines to drive new revenues in new-style search, micro-targeted advertising and big data.

Twitter is the latest to move, acquiring Madbits, a deep learning computer vision start-up. This is part of a race to create the best back end system to deliver brand new user experiences, with the successful companies aiming to define the next generation web interface as Google did with search, and to enable new revenue streams based on intimate knowledge of each user.

These experiences will be deeply context-aware; will support new and intuitive search and query, with voice, gesture and virtual reality interaction; and will provide extremely detailed levels of data about each user, which will then support revenues from analytics and targeted promotions.

IBM's Watson, and the engines behind Apple Siri and Google Now, are among the early movers, while Facebook has been active on the user experience side with its purchase of virtual reality firm Oculus VR. Google (with DeepMind), Yahoo and Dropbox have also made acquisitions in deep learning and machine vision, and Microsoft and Baidu are investing huge sums, so it is no surprise to see Twitter joining the race - initially to support specific new applications in image recognition, but with the potential to create a whole next generation experience.

Its new acquisition, Madbits, has been in stealth mode and was founded by Clement Farabet and Louis-Alexandre Etezad-Heydari, who had worked under Facebook's artificial intelligence director Yann LeCun, also a professor at New York University.

Madbits provides a top level outline of its approach on its website, indicating a focus on understanding and analyzing images. It says: "Over this past year, we've built visual intelligence technology that automatically understands, organizes and extracts relevant information from raw media. Understanding the content of an image, whether or not there are tags associated with that image, is a complex challenge. We developed our technology based on deep learning, an approach to statistical machine learning that involves stacking simple projections to form powerful hierarchical models of a signal."

Image recognition and analysis will help Twitter add image search as well as understanding what its users are tweeting about, and in what context.

The start-up says it has prototyped and tested about 10 different applications, and was on the point of commercial launch when it entered talks with Twitter, "a company that shares our ambitions and vision and will help us scale this technology".

Back to blog