15.2K ·GitHub
What is this?
It is Google DeepMind's research repository: implementations and illustrative code that accompany its published papers, released alongside open-source environments and datasets the community can build on.
It is Google DeepMind's research repository: implementations and illustrative code that accompany its published papers, released alongside open-source environments and datasets the community can build on.Is it one library I can install, or a collection of separate projects?It is a collection of dozens of self-contained projects, one folder per publication — from the Deep Q-Network and Differential Neural Computer to AlphaFold CASP13 and Perceiver IO — not a single installable package.
It is a collection of dozens of self-contained projects, one folder per publication — from the Deep Q-Network and Differential Neural Computer to AlphaFold CASP13 and Perceiver IO — not a single installable package.How do I get the code for a paper I'm interested in?Clone the repository and open the folder named after the paper you want; each project is self-contained, so you follow the instructions inside its own folder.
Clone the repository and open the folder named after the paper you want; each project is self-contained, so you follow the instructions inside its own folder.What do I need on my machine to work with this code?Mostly Python and Jupyter notebooks; the repo ships no unified requirements file, Docker setup or CI, so you install each project's dependencies yourself.
Mostly Python and Jupyter notebooks; the repo ships no unified requirements file, Docker setup or CI, so you install each project's dependencies yourself.Is the code production-ready, or more of a research artifact?It is illustrative research code, not a supported product — completeness varies by project, and the repo explicitly disclaims being an official Google product.
It is illustrative research code, not a supported product — completeness varies by project, and the repo explicitly disclaims being an official Google product.Who is this repo for?Machine-learning researchers, students and engineers who want to reproduce or extend DeepMind's published work; DeepMind explicitly invites the broader research community to engage with it.
Machine-learning researchers, students and engineers who want to reproduce or extend DeepMind's published work; DeepMind explicitly invites the broader research community to engage with it.Can I experiment in the same environments DeepMind uses for its own research?Yes — the repo points to open environments used in DeepMind research, such as DeepMind Lab and StarCraft II, so you can run experiments in the same settings described in the papers.
Yes — the repo points to open environments used in DeepMind research, such as DeepMind Lab and StarCraft II, so you can run experiments in the same settings described in the papers.Does it include famous results like AlphaFold?Yes, the AlphaFold CASP13 implementation is here, along with dozens of other projects: Perceiver IO, PolyGen for 3D mesh generation, RL Unplugged offline-RL benchmarks, and the 2022 Nature work on controlling tokamak plasmas with deep reinforcement learning.
Yes, the AlphaFold CASP13 implementation is here, along with dozens of other projects: Perceiver IO, PolyGen for 3D mesh generation, RL Unplugged offline-RL benchmarks, and the 2022 Nature work on controlling tokamak plasmas with deep reinforcement learning.Why does DeepMind give this code away?The stated goal is to accelerate scientific progress: publishing runnable code, environments and datasets lets other researchers verify, reuse and extend the work instead of starting from the paper alone.
The stated goal is to accelerate scientific progress: publishing runnable code, environments and datasets lets other researchers verify, reuse and extend the work instead of starting from the paper alone.What are similar projects?DeepLearning, google-research, math, maths-cs-ai-compendium, sutskever-30-implementations, mit-deep-learning, spinningup, reinforcement-learning-an-introduction
DeepLearning, google-research, math, maths-cs-ai-compendium, sutskever-30-implementations, mit-deep-learning, spinningup, reinforcement-learning-an-introductionWhich topics is it in?graph-neural-networks, science, physics, jupyter, neural-networks, papers, examples, algorithms, reinforcement-learning, jupyter notebook, academic-research, artificial-intelligence
graph-neural-networks, science, physics, jupyter, neural-networks, papers, examples, algorithms, reinforcement-learning, jupyter notebook, academic-research, artificial-intelligence