Open Source Code


A library of reinforcement learning building blocks.

Blog post • View source on GitHub


A highly-customisable gridworld game engine with some batteries included, useful for discovering and testing reinforcement learning agent capabilities on small problems.

View source on GitHub

Multi-Task Self-Supervised Resnet V2

A vision network trained via large-scale multi-task self-supervised learning

View publication • View source on GitHub

I3D model

Convolutional neural network model for video classification trained on the Kinetics dataset.

View publicationView source on GitHub


A TensorFlow implementation of the Differentiable Neural Computer, which accompanies our Nature paper 'Hybrid computing using a neural network with dynamic external memory'.

View publication • Blog post •  View source on GitHub

DeepMind Sonnet

Sonnet is a high level library to build complex neural network structures in TensorFlow. 

Blog post • View source on GitHub

L2L in TensorFlow

Learning to Learn (L2L) is a TensorFlow implementation of 'Learning to learn by gradient descent by gradient descent', a fully differentiable approach which learns optimization algorithms by optimizing them.

View publication • View source on GitHub


Source code of DQN 3.0, a Lua-based deep reinforcement learning architecture for reproducing the experiments described in our Nature paper 'Human-level control through deep reinforcement learning'.

View publication • View source on GitHub