Workshop 29 11 19

First meeting of the projet fil rouge

Resources :

• Explore the resources sent by Florence
  • AutoML website combines all types of resources in the domaine of AutoML.
  • Databricks' documentation on AutoML
• Something helpful:
  • AutoML/DL Challenge : A collection of series of AutoML challenges (just like the competitions in Kaggle) held since 2015 for reaching the immediate practical interest of the field progresses as milestones. Could it be of interest for us to participate in one of these challenges ? The advantage is that these challenges provide data and problems in details which facilitates the understanding of the abstract conceptions in AutoML. This remains to be discussed with Florence and Nacim. (Note : the organisers are near us Isabelle Guyon, U. Paris-Saclay; UPSud/INRIA, the challenges should be famous here).

Questions for meeting with Florence on 3 dec 12h45 to 13h30

1. Do we focalize on only deep-learning/neutal networks in this project?
2. How to make this project from a large general abstract one to something with details, clear targets that we can deal with?
3. From our limited understanding, NAS is not the only method for hyper-parameters searching, others like Blackbox HPO, do we need to consider ?

Current milestones for the project

1. A presentation of progress needs to be done by the end of Jan 2020
2. The total project ends at June 2020

Github/google drive for the collaboration working

1. Personal notes on the articles/understanding of methods will be collected and published on github - wiki
2. Work with the Git-Flow method to control the version and project progress. On the first stage of project: Etat de l'art, we work on the "documentation" branch and then merge to master.
3. A Kaban is available on the projects page of the repository to store our active and futur tasks during the project.
4. Store the large files (eg. pdf) in this Google Drive shared folder