Information Technology Reference
In-Depth Information
This “winner-take-all” approach has produced recent and important advances in algorithm
design and refinement, but has caused the rejection of many algorithms that has an excellent
performance on an specific cases, but result uncompetitive on average. The following two
questions emerge from the literature (Leyton-Brown, 2003). How to perform an algorithm
selection for a given instance? How to evaluate novel hybrid algorithms?
a.
Algorithms with high average running times can be combined to form a hybrid
algorithm more robust and with low average running time when the algorithm inputs
are sufficiently easy and uncorrelated.
b.
New hybrid algorithm design should find more robust solution and focus on problems
on which a single algorithm performs poorly.
c.
A portfolio of algorithms can also be integrated through the use of hybrid algorithms
because the solutions are considering more innovative.
In previous section we use machine learning algorithms to automatically acquire knowledge
for algorithm selection, leading to a reduced need for experts and a potential improvement
of performance. In general, the algorithm selection problem can be treated via meta-learning
approaches. The results of this approach can cause an important impact on hybridization. In
order to clarify this point, is important to speculate about how the empirical results of meta-
learning can be analyzed from a theoretical perspective with different intentions:
a.
Confirm the sense of the selection rules
b.
Generate insights into algorithm behavior that can be used to refine the algorithms.
The acquired knowledge is confirmed when the performance of the refined algorithms is
evaluated. The knowledge can be used to integrate complementary strategies in a hybrid
algorithm.
6.2 Use of hybridization to solve ASP in social domains
The principal advanced in the reduction of Complexity is related with the amalgam of
different perspectives established on different techniques which to demonstrate their
efficiency in different application domains with good results.
Hybridization of Algorithms is one of the most adequate ways to try to improve and solve
different ASP related with the optimization of time. Many applied ASP´s have an impact on
social domains specially to solve dynamic and complex models related with human
behavior. In (Araiza, 2011) is possible analyze with a Multiagents System the concept of
“Social Isolation”, featuring this behavior on the time according with interchanges related
with a minority and the associated health effects, when this occurs.
In addition, is possible specify the deep and impact of a viral marketing campaign using a
Social Model related with Online Social Networking. In (Azpeitia, 2011), an adequate ASP
determines the way on the future of this campaign and permits analyze the track of this to
understand their best features.
6.3 Future trends on the resolution of ASP using a hybrid system of metaheuristics
We expected that the future trends for solving ASP with hybridization will be based on
models that tend to perform activities according to a selection framework and a dynamic
