DeepMind Open Sources Sonnet Library To Make It Easier To Create Complex Neural Networks

Short Bytes: Sonnet is a new open source library announced by Alphabet’s DeepMind. It is built on top of their existing machine learning library TensorFlow along with extra features that fit DeepMind’s research requirements. Sonnet is designed to make it easier to create complex neural networks using TensorFlow.

DeepMind has already been doing wonders in the open source world. Last year, they open sourced their game-like AI training platform called DeepMind Lab.

Now, they have open-sourced a new library called Sonnet which is designed to ease the neural network construction using TensorFlow. However, it doesn’t replace TensorFlow.

Also Read: Google’s First Machine Learning Chip (TPU) Is 30x Faster Than CPUs And GPUs

Sonnet follows an object-oriented approach, said DeepMind in their blog post. It makes the creation of the modules easier. Modules are the small fragments – assigned with a small task – that are combined to form a neural network.

“Modules are ‘called’ with some input Tensors, which adds ops to the Graph and returns output Tensors.” Adding transparency to variable sharing was accomplished by automatically reusing variables on subsequent calls to the same module.

Changes have been made to TensorFlow to consider models as hierarchies, this would allow switching between modules easier while experimenting, without doing tedious code changes.

DeepMind foresees Sonnet to be used by the community as a research propellant. Also, it would allow easy sharing of other models created by DeepMind with the community.

Sonnet is available on GitHub. DeepMind has published a paper describing the initial version of Sonnet. Also, they have set up a new web page which would be the showcase of their current and future open source projects. Recently, Google also announced a similar online portal for their open source projects.

If you have something to add, drop your thoughts and comments.