add gpu & deep-learning support

[c0mm4nd]

Motivation: support the simulation for the federated learning framework on blockchain
Current: no gpu support (told by Prof. Du on his lecture, SUSTech)
Approach: Docker's native gpu support https://docs.docker.com/compose/gpu-support/

TODO:

• base docker images with pytorch/tensorflow
• pytorch/tensorflow services (with server)
• fed-learning (with blockchain) example

BTW: the deploy.resources field is also a good approach to set the limits and reservations, but I feel hard to write the template, which is holy inflexible. What's your attitude on rewriting the compilers and generating the docker-composer.yml with the PyYAML lib?

[kevin-w-du]

This is very interesting. Once I settled down from my travel next week, let's have a chat using Wechat or other platform. I want to learn more about what you are trying to do, which looks very interesting. Please send an email to me ([email protected]), so I can send you my wechat id.

[kevin-w-du]

So far, whether our emulator can support learning algorithms was asked by the audience in three of my presentations. This is definitely an important feature that we should add to the emulator. Thanks for your work.