1.6. Selection and Adjustment of Grain-Sizes

The ability to select a proper grain-size world for a given problem is fundamental to human intelligence and flexibility. The ability enables us to map the complex world around us into a simple one that is computationally tractable. At the same time, the attributes in question of the world are still preserved. As mentioned before, it is not necessarily true that any classification can achieve the goal.
How to select a proper grain-size is a domain-dependent problem mainly. We'll discuss some general principles below via semi-order spaces.
It has been shown that the compatibility of R and T is the foundation for assuring that image is also a semi-order space. We will discuss below how to adjust R such that it is compatible with T.

1.6.1. Mergence Methods

Assume that image is a semi-order space, where X is a finite set, R is an equivalence relation on X and is incompatible with T. image is a quotient space corresponding to R. We will discuss how to adjust R by mergence methods.
We define a new partition image, where ‘<’ is a topology image on image induced from T. Obviously, image is a partition of X. Let image be an equivalence relation corresponding to partition image.

Proposition 1.9

image defined above is compatible with T. If an equivalence relation image satisfies (1) image and T are compatible, (2) if image, then image, i.e., image is the finest one, where image image if image , then image.

Proof:

image on image is a right-order topology induced from T. [X] is a quotient space corresponding to R. image is a quotient topology on [X]. image and image are quotient spaces corresponding to image and image, respectively.
For image, if image, from the definitions of image and image, we have image, i.e., any neighborhood of a must contain b and is also the neighborhood of b. Conversely, any neighborhood of b must contain a and is also the neighborhood of a. Thus, the neighborhood systems of a and b are the same. Their common neighborhood system is just the neighborhood system of a′ on image. In other words, if elements on [X] having a common neighborhood system are classified as one category, then [X] is revised as image.
Now we prove that image is compatible. Assuming image and image, from the above discussion, image and image on image have a common neighborhood system. image, image, a and b have a common neighborhood system. From the definition of image, a and b belong to the same category on image. Thus, image, i.e., image is compatible.
Finally, we prove that image is the finest one. Assume that image is any partition and image.
We will prove below that image. Let image be a quotient space corresponding to image and its projection be image. For image and image, let image, image.
Assume that image is any neighborhood of image on image. Since image, regarding image as a set on image, it is open, i.e., image is open on image. Since image, image, i.e., image is a neighborhood of x. On the other hand, since image, x and y have the same neighborhood system, i.e., image is also a neighborhood of y. Thus, image. We have image, i.e., image.
Similarly, image, i.e., image.
Finally, image.
..................Content has been hidden....................

You can't read the all page of ebook, please click here login for view all page.
Reset
3.133.133.61