Steering Group Meeting 5 February, 2019 Agenda

From neicext
Jump to navigation Jump to search


NLPL steering group meeting

Time: 2019-02-05 : 10:30 - 12:00 CEST Video connection: https://cscfi.zoom.us/j/9917669499

Place: Video conference

Invited:

  • Tomasz Malkiewicz, NeIC (PO)
  • Joakim Nivre, Uppsala University
  • Jörg Tiedemann, University of Helsinki
  • Martin Matthiesen, CSC-IT Center for Science Ltd
  • Stephan Oepen, University of Oslo (and Uninett Sigma2 AS)
  • Anders Søgaard, University of Copenhagen
  • Filip Ginter, University of Turku
  • Leon Derczynski, IT University, Denmark
  • Bjørn Lindi, NeIC (PM)

NLPL-SG 19-00 Welcome and Attendance

Attending:

  • Tomasz Malkiewicz, NeIC (PO) (Remotely)
  • Joakim Nivre, Uppsala University
  • Jörg Tiedemann, University of Helsinki
  • Martin Matthiesen, CSC-IT Center for Science Ltd (Remotely)
  • Stephan Oepen, University of Oslo (and Uninett Sigma2 AS)
  • Filip Ginter, University of Turku
  • Leon Derczynski, IT University, Denmark
  • Bjørn Lindi, NeIC (PM)

NLPL-SG 19-01 The Agenda for the SG meeting

  • 10:30 NLPL SG 19-00 Attendance and agenda (1')
  • 10:31 NLPL SG 19-01 Approval of the agenda (1')
  • 10:32 NLPL SG 19-02 NLPL and Dellingr (John White, Dellingr Project Manager 20')
  • 10:52 NLPL SG 19-03 Status of the Project / Virtual Laboratory (10')
  • 11:02 NLPL SG 19-04 Summary of staff effort (5')
  • 11:07 NLPL SG 19-05 Summary of computing resource in 2018 ('10)
  • 11:17 NLPL SG 19-06 Transfer of the Project Results / Sustainability beyond 2019 (15')
  • 11:32 NLPL SG 19-07 Expectations towards NeIC and national providers (10')
  • 11:42 NLPL SG 19-08 Review of project plan for 2019 (10')
  • 11:52 NLPL SG 19-09 Next Meeting (5')
  • 11:57 NLPL SG 19-10 Any Other Business (3')

NLPL SG 19-02 NLPL and Dellingr

Dellingr:

  • First pilot with resource sharing by exchanging CPU-core hours.
  • Second pilot using Waldur software to automate processes
    • implications for cost as there is different policies in different countries
    • NLPL is a use case. Input has been provide by Stephan and Bjoern, but no practical use so far
    • MAX4 will be first use of Waldur
      • Dellingr will learn from it
        • Website: shared.dot.neic.no
  • Waldur on Fram or abel, issues on taito. All systems uses SLURM - technical feasible, though billing units vs CPU-core hours
  • NeIC/Dellingr will not provide CPU hours for projects like NLPL. National providers are access points.
  • Discussed NLPL continuation:
    • administrative mechanisms for computing/storage for the project
    • ability to work as a group,the need to receive an allocation as a community.
    • policy of the national exchange of resources
      • high level track
      • directors (not Dellingr, not NeIC)
  • Meetings with Directors in February: mention concern from NLPL regarding sustainability

NLPL SG 19-03 Status of the Project / Virtual Laboratory

All major milestones are reached. The Virtual laboratory is taking shape. There is increasing use and feedback. Public repositories are gaining visibility. Associate sites are using the resources. UiO and UU use the virtual laboratory for master classes

NLPL SG 19-04 Summary of staff effort

Some room for maneuvering - the efforts from CSC and USIT is almost not used. Three iterations of the spreadsheet.

NLPL SG 19-05 Summary of computing resource in 2018

There is increasing use of the virtual laboratory. Overall usage is substantially up. There is a divisioning of labor between taito and abel. Students use and associates use Abel. Access to these systems are very important for non-Finnish/Norwegians. UU and UiO use the NLPL environment on abel for master level courses.

NLPL SG 19-06 Transfer of the Project Results / Sustainability beyond 2019

If all NeIC-funded parts are removed from the NLPL-projects, what remains is very close to a running project:

- A "steering group" with representatives from the project partners
- A software and data collection on two different systems
- A wiki describing the project

All project representatives are eligible for applying for compute time in Finland, Sweden, Denmark and Norway through either NeIC-Dellingr or the national providers. NLPL can formulate a requirement to NeIC-Dellingr and the national providers emphasizing the importance of cross-boarder access, that this is something the project expect is doable.

What is not present is a project manager and funding for the winter school. The Winter School can be continued if each partner raises 2000-3000 Euros. Project Management can be done a rolling role as sketch in the table below:

Rolling Project Managers
year 2020 2021 2022 2023 2024 2025 2026
Project Manager UU UT UH ITU CU UiO
Project Assistant UT UH ITU CU UiO UU

Project goals can formulated and administered through a web based kanban board: CodeRefinery's kanban board


  • Things to protect
    • Policy Framework
    • Virtual laboratory
      • Data stack - Couple of TB of data
      • Software stack
      • Mechanisms for backup, replication and accounting
    • External Services
      • OPUS
      • Vector Repositories
    • Winter school
    • Communities
  • CLARIN community?
    • Collaborating/pooling resources
      • Varies among the countries

Discussion: NeIC: The suggestion is in line with mandate/strategy. The benefits realization needs input. UU: Is positive to the proposal. The key issue is resource sharing. How can this be achieved? CSC: Like the idea. Could CLARIN be involved? As a multinational structure it could possibly replace NeIC. In Finland CLARIN is applying for funding. The application could mentioned support for NLPL as part of the language bank of Finland. UiH: Winter School is important UiO: What is valuable to us? See 'Things to protect' Minutes from NeIC Board meeting from June mention that the small projects are administrative heavy. Could there be alternatives, like financial contribution to a winter school? NeIC: Not likely as Nordforsk has some experience with light weight project(s) with a 5 year commitment. This ended '18 and the conclusion is that was not very successful. UiO: NeIC and national providers need to discuss how they can help NLPL into sustainability. NeIC: NeIC will think about how to deal with alumni. Needs to elaborate on how to formulate expectations. UiO: The project needs to formulate expectations, provide very elaborate requirements. Seems NeIC want NLPL to take requirements to national providers.

Conclusion of discussion:

    NLPL should write a summary listing the what is valuable.
           - List of open questions
           - List of what NLPL want to protect

NLPL SG 19-07 Expectations towards NeIC and national providers

Should the project use resources to establish one system/virtual laboratory in each country? Not especially efficient if every project must balance between all Nordic countries. We should assume resource sharing at a national level, above the project level is in place. (Mentioned as part of the discussion of the project plan)

  • Not cost-efficient to have 4 NLPL installation in 4 Nordic countries
    • 1-2 Nordic countries should be able to serve the community

The project must expect that access to resources is dealt with at a administrative level, such that the NLPL can apply for resources as group. Further must NLPL assume that at the technical level tools are in place to administer access and resource sharing. Waldur can be installed on Taito and Abel/C1 to accommodate this.

NLPL SG 19-08 Review of project plan for 2019

How can I train (https://medium.com/syncedreview/best-nlp-model-ever-google-bert-sets-new-standards-in-11-language-tasks-4a2a189bc155 BERT)? Only Google are capable. There is a methodology component which should be addressed in NLPL. Is there an interest in proposing a workshop for NoDaLiDa? The Winter School is a good opportunity to discuss the topic.

All partners should review (https://wiki.neic.no/wiki/File:20180515_NeIC_NLPL_projectplan.pdf the project plan). Some milestones might need to be changed since NLPL if will be without support from NeIC from 2020.

UiO invited to participate in the pilot testing of the new system C1 in Norway. There is also replacements on its way in Finland. These new systems will also influence how NLPL spend its resources this year.

NLPL SG 19-09 Next Meeting

  • Wednesday, May 29th, 2019, 13:00-14:30 CET

NLPL SG 19-10 Any Other Business