|
One of the characteristics of the evolutionary multi-objective approach is that it
takes significant time to converge. So, we need algorithms that can converge in
reasonable amount of time. In general, the greedy algorithms are good candidates
when it comes to fast execution. Ordinary greedy algorithms do not use dominance
as criteria of deciding the best among individuals. So, we use dominance
to drive the greedy decisions. Also, greedy algorithms generate one solution. But,
in our system, the notion of conflicts between objectives implies that there is no
one best solution, so we use a multiple random starting points to generate different
solutions. A pseudo code is listed below for the multi-objective randomized
greedy approach. The star network configuration behind using this algorithm is its fast execution
time which could be necessary sometimes if we need to generate quick solutions
when the network is not in a good shape. The uniqueness of this algorithm relies
in the several starting points that give us a variety of solutions and relies
in the use of dominance as criteria of comparing solutions. The first for loop
of the algorithm generates different starting points to generate solutions from.
Each starting point is basically a node in the system. In each starting point,
the algorithm considers replicating on neighbors. Here, two cases may happen, if
replicating on any of the neighbors does not dominate the current solution, then
the current solution is the final solution. Otherwise, we replicate on one of the
neighbors which dominate the current solution and the process will be repeated
from that neighbor until no further improvement can be found.
|
