Information Technology Reference
In-Depth Information
seven algorithms: 14.2%. For the instances of the remaining percentage (100-76%), the
selected algorithms generate a solution close to the optimal.
The selection system with feedback was implemented using a neural network, particularly
the Self-Organizing Map (SOM) of Kohonen available in Matlab 7.0. The best results were
obtained with only two problem characteristic indices (
p
,
t
) in a multi-network. The accuracy
increased from 78.8% in 100 epochs up to 100% in 20,000 epochs. These percentages
correspond to the network with initial-training and training-with-feedback, respectively.
The SOM was gradually feedback with all the available instances. Using all indices (
p
,
b
,
t
,
f
,
d
)
the SOM only reached 76.6% even with feedback.
5.2 Selection of heuristics in a hyper-heuristic framework
A hyper-heuristic is an automated methodology for selecting heuristics to solve hard
computational search problems (Burke et al., 2009; Burke et al., 2010; Duarte et al., 2007). Its
methodology is form by a high-level algorithm that, given a particular problem instance and
a number of low-level heuristics or metaheuristic, can select and apply an appropriate low-
level heuristic or metaheuristic at each decision step. These procedures on their way to work
raise the generality at which search strategy can operate. General scheme for design a hyper-
heuristic is shown in Figure 7
.
E
v
a
l
u
a
t
i
o
n
F
u
n
c
t
i
o
n
D
o
m
a
i
n
Problem to be
solved
Set of low level heuristic or
Metaheuristic
Hyper-
Heuristic
B
a
r
r
i
e
r
H1
H2
H3
Hn
The problem solution space
Fig. 7. Hyper-heuristic Elements
The first low-level algorithms build a solution incrementally; starting with an empty
solution with the goal is to intelligently select the next construction heuristics or
metaheuristic to gradually build a complete solution (Garrido, & Castro, 2009).
5.2.1 Representative examples
SQRP is the problem of locating information in a network based on a query formed by
keywords. The goal of SQRP is to determine the shortest paths from a node that issues a
query to nodes that can appropriately answer it (by providing the requested information).
Each query traverses the network, moving from the initiating node to a neighboring node
and then to a neighbor of a neighbor and so forth, until it locates the requested resource or
