Skip to content

Minutes MAP Open source product committee #3 13 01 2021

Clément Mayer edited this page Jan 18, 2021 · 3 revisions

Thank you to everybody for joining the session and sharing your thoughts and ressources!

Participants

Cédric Meurée (AiVision) Céline Jacques (Apricity) Chris-Alexandre Pena (Apricity) Jérôme Chambost (Apricity) Clément Mayer (Substra Foundation) Eric Boniface (Substra Foundation) Nathanaël Cretin (Substra Foundation) Camille Marini (Owkin / Substra Foundation) Romain Goussault (Owkin / Substra Foundation)

Minutes

Product organization

Today we have 2 different meetings with an overlapping scope (and people). We will set up from February a single monthly meeting that addresses the different topics according to the current events of the moment and the topics.

Last releases

In addition to the various features indicated in the release notes, version 2 of Hyperledger Fabric has been implemented.

Features review

Models distillation

Following the various studies carried out, it would be potentially interesting to provide a distillation library in the SDK.

Unsupervised learning

A concrete use case has been identified in the context of setting up a consortium. The effort to set up unsupervised training should not be substantial when necessary.

End-to-end genealogy

As a result of the efforts that have been made within the framework of Data For Good, a template has been proposed and made available. It is now necessary to identify what is present in the framework to see how to feed the template (automatically?) with the available information (Action Clément / Nathanaël)

Data preview

The idea is not necessarily show data but at least have different statistics and the type of data. Due to privacy problems, it is possible that not all the information is always available (for example, the minimum / maximum). Today this information is available through discussion but data owners do not always know how to provide this information or how to calculate it. Automation would ensure the quality of the information provided. This feature is highly demanded by users. Action Jérôme: list the interesting information to be provided

Documentation

A cheat sheet will be set up per version (because of the difficulty to manage versioning in the doc).

Other questions

  • Which ports should be exposed? Need for clarification on the subject (Actions Clément / Nathanaël)
  • Question: retrieving the weight of a learning: today we have the result but not the model. It is in the CLI today --> The possibility is being investigated for Q1 2021 as part of the Healthchain project. However, privacy issues will have to be resolved on the subject of allowing permissions.