Information Technology Reference
In-Depth Information
3.1 Bioinformatics
In (Nascimento et al., 2009) the authors investigate the performance of clustering algorithms
on gene expression data, by extracting rules that relate the characteristics of the data sets of
gene expression to the performance achieved by the algorithms. This represents a first
attempt to solve the problem of choosing the best cluster algorithm with independence of
gene expression data. In general, the choice of algorithms is basically driven by the
familiarity of biological experts to the algorithm, rather than the characteristics of the
algorithms themselves and of the data. In particular, the bioinformatics community has not
reached consensus on which method should be preferably used. This work is directly
derived from the Meta-Learning framework, originally proposed to support algorithm
selection for classification and regression problems. However, Meta-Learning has been
extended to other domains of application, e.g. to select algorithms for time series
forecasting, to support the design of planning systems, to analyze the performance of meta-
heuristics for optimization problems. Meta-Learning can be defined by considering four
aspects: (a) the problem space, P, (b) the meta-feature space, F, (c) the algorithm space, A
and (d) a performance metric, Y. As final remark, authors demonstrated that the rule-based
ensemble classifier presented the most accuracy rates in predicting the best clustering
algorithms for gene expression data sets. Besides, the set of extracted rules for the selection
of clustering algorithms, using an inductive decision tree algorithm, gave some interesting
guidelines for choosing the right method.
3.2 WEB services
In recent years, many studies have focused on developing feasible mechanisms to select
appropriate services from service systems in order to improve performance and efficiency.
However, these traditional methods do not provide effective guidance to users and, with
regard to ubiquitous computing, the services need to be context-aware. In consequence, the
work achieved by (Cai et al., 2009) proposed a novel service selection algorithm based on
Artificial Neural Network (ANN) for ubiquitous computing environment. This method can
exactly choose a most appropriate service from many service providers, due to the earlier
information of the cooperation between the devices. Among the elements that exist in the
definition of a service,
Z
represents the evaluation value of respective service providers'
service quality, and its value is calculated with a function that involves the time and the
conditions of current context environment, e.g. user context, computing context, physical
context, with a division into static and dynamical information.
Among the advantages of using ANN to solve the service selection problem, is that, the
method can easily adapt the evaluation process to the varying context information, and
hence, it can provide effective guidance so that lots of invalid selecting processes can be
avoided. The neural network selected was Back Propagation (BP) because is the most
commonly used; however, this algorithm was improved with a three-term approach:
learning rate, momentum factor and proportional factor. The efficiency of such algorithm
was obtained because adding the proportional factor enhanced the convergence speed and
stability. In conclusion, the authors claim, that this novel service selection outperforms the
traditional service selection scheme.
