Posredujem vabilo na predavanje na Kemijskem inštitutu. Študija je bila narejena v R-ju in bo morda koga zanimam kak implementacijski detajl. Več pa spodaj.
lp,
Roman
On Uncensored Global Stochastic Optimization in / and
Multi-Walk Algorithms under Problem-Specific Tableau Formulations
Dr. Franc Brglez
Department of Computer Science
North Carolina State University
Raleigh, NC, USA
torek, 18.9.2018, ob 13:00
Velika predavalnica,
Kemijski Inštitut,
Hajdrihova 19, Ljubljana
There are numerous stochastic optimization algorithms in both the continuous domain
R^p and the discrete domain D^p . We use the generic symbol D^p to denote domains
where coordinates can be binary strings in [0,1]^p , ternary strings in [0,1,2]^p,
permutations of a p-tuple [1,2,3,...,p], concatenations of binary and ternary coordinates,
separated by a '.', etc. While graph problems such as the minimum vertex cover are
frequently solved without the use of binary coordinates, the efficiency of the proposed
multi-walk algorithm is significantly improved when we recursively:
(1) implement the multi-walk as tracking the movement of a set of p bots defined by
coordinates in D^p ;
(2) compute, for each bot, a distance=1 neighborhood of r <= p adjacent
coordinates; and
(3) devise an efficient method not only to probe the objective function at each
of p bot coordinates but also an effective method to determine the best coordinate
for the next step of each bot.
We consider problems and solutions in both the continuous and the discrete domain.
In R^p , we examine a number of standard hard instances where neighborhood radius is
defined by tableaus of coordinate differences. In D^p, we consider tableau-based
solutions defined by binary coordinates as well as permutation coordinates. For
coordinates in [0,1]^p, we outline tableau formulations for set cover, set packing, graph
vertex cover and maximum clique, Golomb ruler, and lights-out puzzle. For permutation
coordinates in [1,2,3,..,p], we examine tableau-based solutions for the linear ordering
problems, including the instances of Paley tournament graphs that are progressively
increasing in size.
--
In God we trust, all others bring data.