...

All meeting notes

2024-19-01 (Tuesday) UpcomingUpcoming 

Recording link: https://zoom.us/rec/share/lMXR7SSB3r9MBuHcJ12PvrU3YreTJAfwjk-NwivW4nRK9NyJHL1-eGLBTepeZIa5.iC04E57oIKb1efjn 

• Topics:
  • Dev Sprint 1 2 :   https://lf-o-ran-sc.atlassian.net/jira/software/c/projects/AIMLFW/boards/38/backlog  (19th Aug - 9 sep)
  • Bugs/feature (Recent merge /WIP)
    • Change in flow for 1. training 2. model registration
    • Model registration persisted in postgress
    • https://gerrit.o-ran-sc.org/r/c/aiml-fw/athp/sdk/feature-store/+/13287 (on-going)
    • https://gerrit.o-ran-sc.org/r/c/aiml-fw/awmf/modelmgmtservice/+/13286 (on-going)
  • Model Performance measurement

    Jira Legacy
    server System Jira serverId 3122c0e4-6090-3a7d-9725-8b5a32a8eaeb key AIMLFW-91

    • New repository : 
      • Requested:
        • https://lists.o-ran-sc.org/g/toc/message/1061
      • Created :
        • aiml-fw/apm/analysis-module
        • aiml-fw/apm/monitoring-agent
        • aiml-fw/apm/monitoring-server
  • Integration of AIMLFW with NonRT RIC & F2F demo preparation : Release K: Creating & running an ML-based rApp
  • Import the code related to UAV path prediction use case to repository 
    Demo
    • Issue while data extraction details already shared over slack
    • updated the pipeline based on kfp v2.2.2
    • version code management.
    • Presentation from Geon Kim .(15min)

  Action Item:

  • Identify the AIMLFW expectation from integration test [any use case]
  • Pairwise testing plan

  ...

  • Topics:
    • Dev Sprint 1 : https://lf-o-ran-sc.atlassian.net/jira/software/c/projects/AIMLFW/boards/38/backlog  (19th Aug - 9 sep)
    • Model Performance measurement

      Jira Legacy
      server System Jira serverId 3122c0e4-6090-3a7d-9725-8b5a32a8eaeb key AIMLFW-91

    • Bugs/feature (Recent merge /WIP
      • https://gerrit.o-ran-sc.org/r/c/aiml-fw/awmf/modelmgmtservice/+/13259
    • Support asked: 

      • Jira Legacy
        server System Jira serverId 3122c0e4-6090-3a7d-9725-8b5a32a8eaeb key AIMLFW-109

    • Integration of AIMLFW with NonRT RIC
    • First time contributor
      • Jiwon Ock

  Action Item:

  • Identify the AIMLFW expectation from integration test [any use case]
  • Pairwise testing plan
  • Repository needed for model performance – 
  • update comments from last meeting on model performance [Subhash]

  ...

  • Topics:
    • Model Performance measurement

      Jira Legacy
      server System Jira serverId 3122c0e4-6090-3a7d-9725-8b5a32a8eaeb key AIMLFW-91

    • Bugs
    • Support asked: 

      • Jira Legacy
        server System Jira serverId 3122c0e4-6090-3a7d-9725-8b5a32a8eaeb key AIMLFW-109

    • Integration of AIMLFW with NonRT RIC
    • First time contributor
      • JunHyuk Lim
      • GyuYong Park
      • Kim Do Young
      • inni
      • Sumin Im

  ...

  • Joseph Thaliath presented the I release plan. Summary and more details are there in the "I" Release page.
  • Joseph Thaliath mentioned that if anyone is interested to discuss on those topics or willing to contribute, please send an email to jo.thaliath@samsung.com
  • @Navaneethan asked about the pre-processing step that is required before training the model. How is it handled or it is assumed it will not be there ?
  • Joseph Thaliath mentioned that in the current framework it can be handled as part of training pipeline or limited instructions for data transformation can be given to data extraction component. But it is not clear how it is handled from O-RAN alliance point of view.
  • @Navaneethan asked if  we are storing the model in the pickle format or any other form recommended?
  • Joseph Thaliath mentioned that the framework is transparent to the format, any format, the model sdk zip the file/folder in Model DB. We are assuming the entity using the model knows how to use the model after unziping
  • Deeepanshu Karnwal asked if continuous training flow is supported?
  • Joseph Thaliath mentioned that the current framework supports only one time offline training and if training is required again, it needs to be triggered again using the training API. There is no option to currently use previous model for further training.
  • SANDEEP KUMAR JAISAWAL mentioned that the current framework has option to trigger training API again and again but the only thing that we are not doing is retaining as soon as the new set of data is available.
  • John Keeney (Ericsson EST) mentioned that there will some changes in the interface used by AIMLFW for the H-release maintenance
  • Kevin Timoney will clarify the changes in the interface used by AIMLFW  by email.
  • Joseph Thaliath will look at these changes and see if it can be integrated easily and if there are any queries will contact Kevin or John.

  ...