Changes between Version 1 and Version 2 of FutureProjects


Ignore:
Timestamp:
Apr 16, 2010 4:37:03 PM (10 years ago)
Author:
tkr
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • FutureProjects

    v1 v2  
    1 == Half-baked Ideas ==
     1= Half-baked Ideas =
    22
    3 = Looking at Osi =
     3== Looking at Osi ==
    44
    55Branching information objects exist in Osi. COIN-OR solvers (e.g., CBC, BCP, etc.) having similiar information should have similiar results (e.g., size of search tree, etc.) According to JJHF and LL in their work, this was not the case. Someone should/could/would explore what's happening.  Buzz on the grapevine was that maybe Andreas W. (IPOPT) or Pierre B. (Bonmin) would have interest.
    66
    7 = Auto Tuning =
     7== Auto Tuning ==
    88
    99There are many parameters in the code, and by selecting the appropriate magic values, the code's performance can be greatly changed.  The milestone is to (1) improve Cbc's default values, and (2) have a good set of parameter values be automatically determined by the software itself.
    1010
    11 = Heuristics =
     11== Heuristics ==
    1212
    1313 * '''Relaxation Induced Neighborhood Search''': As I understand it, the RINS heuristic solvers a dozen of so LPs, identifies integer vars that have remained fixed, fixes those integer vars at their sticking point values then branches a while. 
     
    1515 * '''Feasibility Pump''': The initial algorithm was proposed by Andrea Lodi. Rumor has it that variations on it by JJHF perform even better.
    1616
    17 = Preprocessing =
     17== Preprocessing ==
    1818
    1919Wouldn't it be nice if structure in a problem (that could be exploited to more quickly solve the problem) was labeled, so you didn't have to waste processing time finding it?  That's the general idea.
    2020
    21 = Cliques =
     21== Cliques ==
    2222
    2323It is known that at least one illustrious mip solvers of the past successfully made more use of "cliques" (e.g., knapsack cover cuts with cliques.)
    2424
    25 = SWATH Harvest =
     25== SWATH Harvest ==
    2626
    2727In 1999, JJHF presented a solution to the unsolved SWATH problem of the miplib test set (http://dimacs.rutgers.edu/Workshops/Gomory/abstracts.html#forrest).  There were two key ideas.