Article Preview
TopIntroduction
Constructs of Computational Intelligence (CI) (Angelov et al., 2008; Crespo & Weber, 2005; Kacprzyk & Zadrozny, 2005; Kilic et al., 2007; Molina et al., 2006; Pedrycz & Gomide, 1998; Pham & Castellani, 2006; Wang et al., 2009) exhibits a surprising diversity of design methodologies. The concepts and architectures of neurofuzzy systems, evolutionary fuzzy systems are becoming more visible and widespread in the literature.
In spite of this variety, there is a single very visible development aspect that cuts across the entire field of CI and fuzzy modeling, in particular. In a nutshell, such constructs are built around a single data set. What also becomes more apparent nowadays is a tendency of modeling a variety of distributed systems or phenomena, in which there are separate data sets, quite often quite remote in terms of location or distant in time. The same complex phenomenon could be perceived and modeled using different data sets collected individually and usually not shared. The data might be expressed in different feature spaces as the views at the process could be secured from different perspectives. The models developed individually could be treated as a multitude of sources of knowledge. Along with the individual design of fuzzy models, it could be beneficial to share sources of knowledge (models), reconcile findings, collaborate with intent of forming a model, which might offer a global, unified, comprehensive and holistic view at the underlying phenomenon. Under these circumstances an effective way of knowledge sharing and reconciliation through a sound communication platform becomes of paramount relevance, see Figure 1.
Figure 1. A General platform of knowledge reconciliation and collaboration in fuzzy modeling
A situation portrayed in Figure 1 is shown in a somewhat general way not moving into the details. It is essential to note that the mechanisms of collaboration and reconciliation are realized through passing information granules rather than detailed numeric entities.
The general category of fuzzy models under investigation embrace models described as a family of pairs <Ri, fi>, i=1, 2, …,c. In essence, these pairs can be sought as concise representations of rules with Ri forming the condition part of the i-th rule and fi standing in the corresponding conclusion part. It is beneficial to emphasize that in such rules, we admit a genuine diversity of the local models formalized by fi. From the modeling perspective, the expression fi(x,ai) could be literally any modeling construct, namely
- •
Fuzzy set,
- •
Linear or nonlinear regression function,
- •
Difference or differential equation,
- •
Finite state machine,
- •
Neural network
One can cast the fuzzy models in a certain perspective by noting that by determining a collection of information granules (fuzzy sets) Ri, one establishes a certain view at the system/phenomenon. Subsequently, the conclusion parts (fi) are implied by the information granules and their detailed determination is realized once Ri have been fixed or further adjusted (refined).