Advanced permission

[ClementMayer]

Regime that covers serial and parallel learning - empowering the data controller - technique and concept - data governance platform

[ClementMayer]

It is already possible to set up different rights on assets in Substra, but a redesign is necessary to integrate finer cases adapted to all use cases of the framework.
For example, here are some examples:

• All the models trained on a dataset that I made accessible are accessible to me.
• All the models created from an algorithm I provided are accessible to me
• and so on.
  An in-depth reflection must be initiated to think about these different regimes and how to set them up.
  A working group will be organized to reflect on this subject.

[SaboniAmine]

Maybe the Biscuit framework could be the tool that lets you declare permissions per users/roles, in a distributed mode.

[ClementMayer]

Update from last MAP committee (10/09/2020):

• [Chris, Céline] Are there ongoing developments on that aspect? What concepts or approaches of permissions are considered?
• [Amine] An idea: we could imagine permissions (e.g. granted by dataset providers to data scientist) conditional to a privacy budget (differential privacy epsilon)
• [Camille] There is no new development at this stage. Owkin will work on it probably on Q4.

[RomainGoussault]

Closing stale issue.