Google DeepMind open sources Sonnet so you can build neural networks in TensorFlow even quicker

Google’s DeepMind announced today that it was open sourcing Sonnet, its object-oriented neural network library. Sonnet doesn’t replace TensorFlow, it’s simply a higher-level library that meshes well with DeepMind’s internal best-practices for research.

Specifically, DeepMind says in its blog post that the library is optimized to make it easier to switch between different models when conducting experiments so that engineers don’t have to upend their entire projects. To this avail, the team made changes to TensorFlow to make it easier to consider models as hierarchies. DeepMind also added transparency to variable sharing.

It’s in DeepMind’s own interest to open source Sonnet. If the community becomes acquainted with DeepMind’s internal libraries, it will become easier for the group to release models side-by-side with papers. Inversely, it also means the machine intelligence community can more feasibly contribute back by employing Sonnet in their own work.

DeepMind has been particularly active in open source in recent months. It is developing an open-source API to enable research to be done on StarCraft II. Back in December, the team released DeepMind Lab to catalyze generalized AI research — much like OpenAI’s Universe. Open-source projects even got their own home on the DeepMind website.

The library is available now on GitHub. The group plans to continue to release updates as it modifies its library internally.