Several question for HPE support

[Ultimate-Storm]

• Licensing access mechanism? Does the pattern copy blockchain like Ethereum or does it follow the SPIRE Federation mechanism?
• Preconditions for node access, such as if new hospitals want to access. Could they directly access the latest model or they need to contribute some data first(proof of work)
• How much freedom the client/user has. What are the model parameter averaging mechanisms? In addition to weighted average, direct average. Can this part be customized by users? Regarding data or model consistency, can users be trained with different models? (Neural network /SVM)

--User can transform any Keras or PyTorch based ML program that is written using Python3 into a Swarm Learning ML program by making a few simple changes to the model training code by including the SwarmCallback API.

• How could Swarm Learning achive better results then merged model.

• Overfitting? Or because the differece between locally trained dataset is huge, that there’s little feature collapse?

• Order of weight synchronizing? The batch end policy is set the same on all nodes but training speed/progress are different. Who will trigger the synchronizing?

• What's the meaning of callback on epochend? since the batchend will take care of everything and epochend will slow down the progress.