Différences

Ci-dessous, les différences entre deux révisions de la page.

--- groupe-dfo-bbo:acteurs:intern:amal-flimine:reunion-2022-04-26 [2022/04/27 13:35]
flimamal
+++ groupe-dfo-bbo:acteurs:intern:amal-flimine:reunion-2022-04-26 [2022/04/28 13:12] (Version actuelle)
flimamal
@@ Ligne 1: / Ligne 1: @@
 ===== Meeting 2022-04-26 =====
-Title : **Define track**
+Title : **Discussion about Speculative search in Nomad**
+With [[groupe-dfo-bbo:acteurs:profs:sebastien-le-digabel:main-page|Sébastien Le Digabel]]
-Discussion [[groupe-dfo-bbo:acteurs:profs:sebastien-le-digabel:main-page|Sébastien Le Digabel]] about Speculative search in Nomad:
   * Speculative search uses fixed parameters (for each iteration of search) BASEFACTOR = 4.0 (stepLength) and SEARCHMAX = 1 (number of points generated along the direction).
   * BASEFACTOR choice explained (opportunistic search in CS + Mesh factor in MADS).
-  * MADS algorithm iterates on Search and Poll (if search step fails). If search step succeeds, MADS skips Poll and iterates directly on search (especially speculative method which in this case extends to a "line search") until termination criteria.
+  * MADS algorithm iterates on Search and Poll (if search step fails). If search step succeeds, MADS skips Poll and iterates directly on search (especially speculative method which in this case extends to a "line-search") until termination criteria.
-  * Interesting work on new linesearch-based algorithmic framework in DFO/BBO context
   ToDo:
+  * Solving problems with Runner (no generation of a performance profile).
+  * Why previous benchmark of MADS w/out Speculative search leads to indication that MADS is less efficient with this latter?
+  * Reading the "Italians" work on new line-search based algorithms in DFO/BBO context.
+  * Checking MADS successive iterations on Speculative search (especially during early phases).
+  * Conducting sensitivity analysis of MADS to the aforementioned parameters (BASEFACTOR, SEARCHMAX).