Implement POC for Federated Preprocessing Use Case 1 about fedbiomed HOT 4 OPEN

Issue with SecAgg: it fails at the validation of the encryption factors. It seems that the validation list should be length 1, but it is length 3 :(

from fedbiomed.

Hi @sharkovsky,
Is it working while running normal training without analytics? Maybe it is broken with some changes that have been added recently?

Anyway, validation part should be removed because of security concerns. We couldn't find time work on that but we can have a look at it together, either fix the issue or remove it while we can :)

from fedbiomed.

The issue with SecAgg is finally solved!
Here is a list of things that went wrong:

1. clipping_range and target_range values needed to be set properly. In particular, clipping_range needs to be as high as the largest number to be encrypted, and target_range needs to be in the order of 10^3 times clipping_range in order to obtain an acceptable resolution.
2. The SecAgg contexts for node and researcher were somehow in an invalid state for me. This led to researcher and nodes silently using a different biprime value. The solution was to do a thorough fedbiomed_environment clean and reconfigure everything
3. Finally, a value of round=0 is not valid for the encryption and aggregation functions. It appears that it needs to be at least 1.

from fedbiomed.

Following sprint review discussion:

• rearranged a bit the tasks in this issue to try to loosely reflect order of implementation/priority
• introduced refactor of experiment and training plan
• removed experiment breakpoint because no longer relevant

from fedbiomed.