How OpenMined uses your idle consoles to train the next generation of AI

By now, you should already have heard plenty about the open-source organization OpenMined. They are working on a way to democratize the access to personal data using Blockchain— and the Unity Game Engine.

Never heard of OpenMined? No problem, there are plenty of articles out there. Just pick one.

OpenMined uses the Unity Game Engine for PySyft, their encrypted Deep Learning Library. You might ask yourself now, why a Deep Learning Library uses a Game Engine. Let me explain.

GPU everywhere for everyone

If you are active in the Deep Learning space, you know that training an AI takes a large amount of processing power. In most cases, a good CUDA-compatible GPU can massively speed up the training progress. Even training a simple classifier can be a pain if you don’t own the appropriate hardware.

This is a problem for OpenMined.

OpenMined targets everyday users, not just Data Scientists or Deep Learning Engineers, who probably own a high-end GPU anyway. Think about your parents, your brothers and sisters, your non-tech friends: Do they own CUDA hardware?

If you have read some of the articles linked above, you know that the goal of OpenMined is that users train the shared network with their own data on their own devices— the personal data shouldn’t leave the device. That’s why it is important that the training is possible on devices that most people use.

By using the Unity Game Engine, the OpenMined team is able to cross-compile their code to almost every existing GPU on the planet. This means that the training can be run on a lot more devices — you probably own multiple of them:

• Consoles: The Xbox One X has a GPU that is comparable to a GeForce GTX980. Additionally, consoles sit idle for most of the time anyway — so why not use it to train some AI.
• Smartphones / Tablets: GPUs in phones are getting better and better — and mostly sit idle if you aren’t a big mobile gamer. If you own an iPhone 8/8 Plus or an iPhone X you also have the power of the dedicated neural network processor A11 Bionic on your hands.
• Computers: Even if your computer or laptop does not have the newest NVIDIA GPU inside — Unity makes it possible to train AI on older GPUs and on hardware from AMD, Intel etc. — on every OS.
• More: Your smart TV probably has a GPU inside, maybe your smartwatch has too, and there are more devices coming all the time.

Conclusion

I said it in my last article, and I’m saying it again: The OpenMined project has the vision and the tools to revolutionize the way we collect, protect and secure your personal data and use that for Artificial Intelligence. Obviously, revolutionizing the way we use our hardware for training was a part of this too.

Large parts of this were taken from the “Why Unity?” Markdown file on the OpenMined GitHub. If you want to contribute, check out the website of OpenMined for more infos and join our Slack. I’m going to say hello.