Google opens the source code of its AI tools. A necessity foremost

Google has never made any secret of his ambitions for the development of artificial intelligence and do not hesitate to communicate frequently about the work and findings of its research labs in the field. We were able to discover the image recognition algorithms used by the Mountain View company and psychedelic landscapes wish that they can produce.

The search engine giant announced Monday it opened its bookstore open source TensorFlow, intended for calculation based on the logic of deep learning. This is particularly the tool that Google uses as part of its image recognition tools, automated its responses to emails or functions in its machine translation.

Google lends its toys

“We could see the possibilities TensorFlow and we think it could have an even greater influence, which is why we are making open source. We hope that through this, the community around the machine learning (researchers, engineers and amateurs) will exchange ideas more easily, via functional computer code and not only through academic papers, “said Sundar Pichai, new CEO Google, on the company blog.

The tool is widely described as elastic by Google, which can operate both on a single smartphone in a data center on multiple machines. The version presented by Google remains for the moment confined to a single machine but the company promises to offer shortly variations for those who would like to deploy the Google tool on a more complex infrastructure.

Business Insider notes that Google still does not play the card of the complete opening and retains secrecy around its data centers and infrastructure used internally by the company. Google hopes that the opening of its code will make further progress in the area of ​​machine learning [Editor’s note: machine learning]., A booming sector at present

Facebook has also begun unveil its research activities in this field, which should particularly benefit M, its new project of personal assistant.

The machine learning at Facebook, Criteo and Microsoft

Pour into open source some of its developments, is it an isolated practice in the technology sector? Not for machine learning, as could confirm yesterday at the Sorbonne, at a conference, research directors of three market players (Microsoft / INRIA, Criteo and Facebook).

” Of course there is a certain competition between the companies, especially when it comes to recruiting talent. But today many companies, including Facebook, share the same vision, namely that the problems we seek to fix are infinitely complex. And even if we employ the best talent, it is not enough to solve these problems such as unsupervised learning “explains Florent Perronnin , Director Parisian research laboratory dedicated artificial intelligence.

The social network and boasts a very open search model with a sharing of different publications. Clearly, current practices in the field of scientific research apply.

“The largest conference of machine learning domain hosting many companies. We must not forget that to develop a product, there is a prior research part. It can and should in my view be shared “abounds Nicolas Le Roux, research director at Criteo .

The joint laboratory Microsoft and INRIA is no exception, confirms its director Laurent Massoulié . Besides scientific publications, software developed in this framework is published as Open Source.

Share and Enjoy

  • Facebook
  • Twitter
  • Delicious
  • LinkedIn
  • StumbleUpon
  • Add to favorites
  • Email
  • RSS