google-deepmind/deepmind-research

DeepMind Research is a single GitHub repository where DeepMind, the AI lab now part of Google, posts the code that accompanies its published research papers. It is not one product or one library. It is a folder of many separate sub-projects, each linked to a paper the lab has put out, ranging from reinforcement learning experiments to physics simulation, language modeling, protein structure work, and more. The README frames the purpose plainly: along with publishing papers, DeepMind releases open-source environments, data sets, and code so the wider research community can engage with the work and build on it. The stated goal is to accelerate scientific progress. The README also points to separate DeepMind repositories for well-known systems like the Deep Q-Network, the Differential Neural Computer, the DeepMind Lab 3D environment, and the StarCraft II learning environment. The bulk of the README is a long list of projects, each entry naming a paper and linking to a sub-folder in the repo. Examples include work on controlling tokamak plasmas with reinforcement learning published in Nature 2022, precipitation nowcasting with deep generative models, the Perceiver IO architecture, Bootstrap Your Own Latent (BYOL), graph network physics simulation, and the AlphaFold CASP13 release. The README closes with a disclaimer that this is not an official Google product.