Points to talk:
  * Performance indicators survey !!!!!!!!
  * DMulti-MADS : summary
     - What has been done
       * Look at some specific functions (maybe not so specific as dms fails at most).
       * Test a hack: did not work.
     - Some stuff to do:
       * Try to test COCO platform; quite incomplete and biased (bounded)
       * When people do not know, they start from a simple set of points (line strategy); could try from a latin hypercube sampling. ** For Sebastien: what must convince you to use this method ? **
       * The values of hypervolume are maybe not good ; choice of the reference point, ...
     - Change of tactic? Stop progressive barrier approach and move to a merit function approach ?
     - Help to organize priorities (until the end of the PHD).
  * Others
     - Julia interface: works well; wait for Alexis to create the package.
     - Suggestion: a guide to review articles for the DFO community.