The current AI decision-making method is linear, given a condition x, substituting a curve to obtain a Y directly as the decision value. Slightly more flexible, take two values from the two curves y1,y2, and then randomly get a value from the middle of the area.
However, doing so will give the AI a very clear "boundary", where the AI can communicate with the AI with a noticeable sense. The feeling will be more "stiff". The reason is that the human decision-making process does not have such boundaries. When we face a certain condition X, our decision "curve" should look like this.
, human decision-making does not have a clear border, so it looks very vague. Fuzzy processing seems to be a very common, in the field of AI research more content. I've also been thinking about how to implement this fuzzy decision-making logic recently. It is a recent thought that since it can be expressed as a blurred image, why not use the image directly. The idea is to start with a picture, and for the sake of simplicity, let's assume it's a 32*32 picture.
We assume that the lower-left corner is 0 points, and the horizontal is x vertical y. Read the picture into the computer and read the red color values of each pixel to form a 32*32 matrix. The matrix landscape represents the condition parameter x, which represents the possible probability distribution of the Y-value vertically. This corresponds to each x value and we are able to obtain a sequence number. In this simple example, let's assume x=18, and get a curve like this.
In this curve, the horizontal axis represents the possible Y value, and the ordinate represents a value that is proportional to the probability that the Y value appears, and we would call it F (y). Assuming that the value for summing this series of discrete points on the graph is Σ, then the probability distribution function p (y) of y can be expressed as P (y) =f (y)/σ. If we leave him alone, we will use this F (Y) directly. We stitch together these values and remember each f (Y) corresponds to the Y value.
The pie chart above is a good illustration of this operation, this cake is the scale of the week is Σ is 3009, the right side of the color table indicates the pie above each color area represents the Y value. The rest of the work is running out. We randomly 0~3009 a number between the numbers in a clockwise direction, and you will find that this stochastic process satisfies the P (Y) probability distribution function we are pursuing.
So, through the three processes of slicing, sorting, and random, we get a "fuzzy" Y value from a given condition X. This way is far more scientific than using envelopes to define the upper bound, and I think if this principle is used to build AI's decision-making system, it will make it very flexible, natural and yet regular.
A method to realize fuzzy decision when writing Texas hold ' em ai
