The main focus of the meeting is working on a roadmap for the activities of the FAIR4ML Interest Group. Specifically, the discussion will be focused on the following objectives:
-
Discuss where FAIR should apply to ML, considering the work in other working groups and focusing on gaps, and
-
Define and prioritize cases for new Task Forces and Working Groups
Ultimately, the outcome of this meeting will be a set of concrete actions for the next 6-12 months, including building a community of practice for information sharing about ML and FAIR pertaining to ML.
In order to facilitate the discussion, people from relevant activities, initiatives and projects will be invited to offer their perspective. Such initiatives will include the Pistoia Alliance, ELIXIR, the CLAIRE network and FAIR4HEP among others, but particular effort will be given towards identifying and engaging with other groups as well. Given the hybrid form of the Plenary, some of these perspectives may be delivered as pre-recorded videos.
Collaborative session notes: https://docs.google.com/document/d/1NWxsoQm9uW1NokWWeHNvWyEYtBP9CvfRPZZd...
The meeting will comprise 2 parts;
-
The first part of the session will be dedicated to organizing the FAIR4ML IG as well as having some initial discussions on Task Forces that can be kickstarted. It will also include some targeted talks from relevant projects or activities, to set the overall stage.
-
The second half of the session will be dedicated to discussing the goals and structure of a potential white paper around FAIR for Machine Learning.
Specifically, a tentative agenda is:
-
Introduction(s) (3 min):
-
What is FAIR?
-
Intro to FAIR 4 ML IG, the objectives and structure
-
-
Part 1:
- Talks (12 min in total), focus on FAIR for ML in different contexts:
- DOME recommendations as a facilitator of FAIRness in ML
- NFDI4DataScience - Linked Metadata for (better) Machine/Deep learning
- The role of registries in FAIR ML (title tbc)
-
Discussion & next steps (40 min)
-
Review IG co-chairs
-
Discuss strategies for expanding the group, including engaging with new groups and attracting more people to the Interest Group
-
Proposed frequency of IG-wide calls, channels of communication
-
Looking through the objectives of the IG, what are the key topics that can be readily addressed through short-term task forces
-
- Talks (12 min in total), focus on FAIR for ML in different contexts:
-
Part 2:
-
Determining the primary goal of a FAIR4ML White paper (30 min)
-
Should FAIR address only ML models, and/or also processes, and/or also platforms?
-
How do you think FAIR should be applied to ML? Reuse FAIR for data and software, or aim towards a redefinition?
-
What changes/definitions are most needed? What questions need most discussion?
-
Capture list of people interested in actively working/leading the effort
-
-
-
Wrap-up (5 min)
-
Review of actions
-
Researchers and Data Professionals interested in Machine Learning, and primarily in how the FAIR principles can potentially be interpreted in the context of ML.
Particularly relevant are members of relevant RDA groups with complementary focus, in order to identify potential synergies early on. An initial list of these key groups is:
-
FAIR for Research Software Working Group (now complete, but the RDA entity which will maintain the WG output is the SSC IG)
-
Software Source Code Interest Group
-
FAIR Digital Object Fabric Interest Group
-
FAIR Data Maturity Model Working Group
There is a large amount of FAIR work, both in RDA and elsewhere, initially focused on data and now software and other products but generally not ML models. Some speakers and attendees in this session are those involved in projects where FAIR for ML models is a topic of discussion. Additionally, we presented poster 31b (FAIR principles for ML models - https://doi.org/10.5281/zenodo.4271995) at RDA VP16 to start discussion on this at RDA, with a dedicated BoF session at RDA VP17 (https://www.rd-alliance.org/defining-fair-machine-learning-ml) that aimed to capture the overall perspective on the topic. The discussion around FAIR for Machine Learning continued in further events under different domains; during the FAIR Festival, the efforts of FAIR4ML were presented together with similar initiatives for Software and Workflows. During ESIP 2021 in July 2021, the initiative was presented in the context of “Best Practices for Reusability of Machine Learning Models: Guideline and Specification”, with a first informal Community Call taking place in the same month as well. Finally, we had two BoF sessions (in VP18 and VP19) towards establishing a critical mass of interested parties and drafting an IG Charter document.
Ultimately, the FAIR for Machine Learning Interest Group was formally accepted in September 2022, making this RDA P20 session the kick-off meeting of the IG.
Just starting
FAIR4ML IG page: https://www.rd-alliance.org/groups/fair-machine-learning-fair4ml-ig
Previous plenary sessions: https://www.rd-alliance.org/defining-fair-machine-learning-ml
- 930 reads