"Someone else must say what is clear to everyone."
Like an epigraph, Google / Yandex did not find the author
When building a model of an object, it can be reduced to the representation of a black box with several parameters P ( i ) that affect the output T. For a complex multidimensional object, these can be models of its "section" in different planes / senses.
In the ideal case, to obtain the information necessary for constructing a model, it is necessary to obtain the values of T for all combinations of its parameters, given with a sufficiently small (uniform) step in the range of values allowed for each of the parameters. The more points there are for each parameter (and, accordingly, T values), the more accurately the model can be built. However, in real life, it often turns out that the parameters affect in different ways, and it is wiser to make a step on each of them unevenly. For example, at the beginning of the permissible interval, the parameter has little effect on T, and in the middle or at the end, its influence changes (and even many times) and, therefore, the step on such a parameter should be taken differently.
But at the beginning of building a model, when there is little information about the relationships, the allowable range of values of each parameter is divided with a uniform step. When re-solving the problem of constructing a model of an object that is close or weakly changed to the previously investigated one, the peculiarity of the influence of its parameters can be taken into account in specifying a non-uniform partition of the admissible interval. This minimizes the number of T values required to build the model while maintaining the completeness of the information received about the object used in the model.
Here you should pay attention to the fact that there are objects for which the cost / complexity / duration of obtaining each output value of T is very high. Further, we will focus on just such objects and their models.
The procedure for specifying each new value of the parameter in the process of studying such an object can no longer be trivial (uniform step), it is highly costly, and should take into account, if possible, all the already available information on the behavior of T. Each step / value of the i -th parameter should take into account its current correlation with the behavior of T. This becomes possible if at each step the preliminary model of the object is built anew, and this is justified, given the cost / complexity / duration of obtaining each new value of T.
( ): , / ( ) . / / , () . . , .
, () . () . , - / . - , . , , () , / . - , , , .. - . ( ) / , .
, ( , Z, ) . . , .
T(X) (X) . , , (. ):
T(X) 3- : Xmin, (Xmin + Xmax)/2 Xmax, .. (Xmin, Xmax). .. GIF 14 .
, Z, , . " " , ( Z) , . / ( ), . , , .
Z ( ) , . , (Xmin, Xmax), 3/2 ( , ) . , .
¾ (!) .
, .
( "", ) , , ( ) Z. , , , - .