Anuket Project

2021-11-12 AI/ML for NFV Meeting Minutes

Attendees

Sridhar Rao

Akanksha Singh 

Kuldip Yadav

Kanak Raj

Ildiko



Sl. No.TopicPresenterNotes
0Time Change 

Currently: 1300UTC

Change to: 6AM Pacific - No overlap with other meetings, and many are OK with the change.

Check: Other attendees from West-Coast.

Request Sandra to Change the calendar Invite.

1About Thoth - Status, Roadmap, Direction. 

2Demo - GANs for generating synthetic data.

Team is unable to attend due to 'Heavy Rains' - Moved to Next-Week.

3Kubernetes - ML

Questions to Answer:

  1. ML problems in Cloud-Native - Existing Works.
  2. Dataset related to these problems.
  3. ML Models used by these existing works.
  4. "Novel problem"  - Not yet considered by any existing works - or a better technique to an existing problem.
  5. Implement the solution to the 'Novel problem' OR better technique for existing problem.

When you do 1, 2 and 3 - it will result in Technical Report ONLY

When you do 4 & 5 - You can publish in reputed Conf/Journals.

Connect Kanak and Akanksha to Girish.

Cloud-Native - Observability is kind of 'Default', even Tracing is common. This results in generating lot of Data and analysis of this data is a hot topic. Are there any ML problems ?

4Openstack Inputs

Apart from CNTT "specification" around Openstack - how it should be used - there isn't any.

Openstack - Ironic-based Baremetal cloud, and from there it can be K8S.

Openstack - Zune : Orchestration of Containers.

5ModelSelector - Phase2

Based on Inputs from smcasey

Current State: Ask users a bunch of questions and then based on the answer decide which tool to use.

Proposal: Build (reuse existing) framework that has 'implementations' of all these models, and collect sample data from user (and the problem) and run through this framework (select set of models) and then decide which model to use based on the 'performance-metric'. This will result in better suggestion/estimation of a 'model-to-use'.

This could be used for MaaS.