Botta, Alessio (2008) Automatic context adaptation of fuzzy systems. Advisor: Lazzerini, Prof. Beatrice. Coadvisor: Marcelloni, Prof. Francesco . pp. 173. [IMT PhD Thesis]
Botta_phdtheis.pdf - Published Version
Available under License Creative Commons Attribution No Derivatives.
Download (6MB) | Preview
Among the several applications of fuzzy set theory, fuzzy-rule based systems (FRBSs) have proven to be extremely successful in a wide range of fields, including, for instance, control, classification, regression, and pattern recognition. In particular, FRBSs have raised attention for their twofold nature, i.e., for their ability to handle linguistic concepts and, at the same time, to perform an accurate modeling of input-output relations. Hence, several researchers and practitioners have developed learning algorithms for the automatic identification of FRBSs from real-world data. Such algorithms include the hybridizations of FRBSs with other popular soft computing techniques, i.e., artificial neural networks and evolutionary algorithms. In this framework, context adaptation of fuzzy systems is considered as an emerging paradigm which has been analyzed only in a few works. In a nutshell, context adaptation consists in tuning some of the features of an already existing FRBS, so as to adapt it to a new configuration of the external environment. This task has usually been approached in the literature as scaling fuzzy sets from a universe of discourse to another. The basic idea presented here is to achieve context adaptation by exploiting a set of operators which allow performing a more flexible tuning than scaling-functionbased techniques, while keeping the semantics and the interpretability of the original FRBS unaltered. Nonetheless, this work collects previous approaches and organizes them in a common framework, thus providing a reference study on the topic. In the development of the thesis, we first recall fundamentals of fuzzy set theory, fuzzy logic, and fuzzy systems. We survey previous work on related subjects and introduce the context adaptation problem in detail. Second, we develop a novel context adaptation approach. To this aim, we introduce a flexible non linear scaling function and four orthogonal fuzzy modifiers which allow adapting an FRBS to any context. Since the modeling capabilities of the operators may negatively affect the semantics of the FRBS, we study two novel indices to properly measure interpretability and prevent such degradation. The proposed learning approach is based on evolutionary algorithms and takes both the flexibility introduced by the operators and the interpretability assessed by the indices into account. We test our context adaptation technique on four different data sets, providing detailed examples and comparisons. Finally, we draw concluding remarks and we discuss future extensions and possible research lines.
|Item Type:||IMT PhD Thesis|
|Subjects:||Q Science > QA Mathematics > QA75 Electronic computers. Computer science|
|PhD Course:||Computer Science and Engineering|
|Date Deposited:||05 Jul 2012 13:28|
Actions (login required)