C++ Neural Networks and Fuzzy Logic: Preface

Weight Determination

You may be wondering about how these weight matrices were developed in the previous example, since so far

we’ve only discussed how the network does its job, and how to implement the model. You have learned that

the choice of weight matrix is not necessarily unique. But you want to be assured that there is some

established way besides trial and error, in which to construct a weight matrix. You can go about this in the

following way.

Binary to Bipolar Mapping

Let’s look at the previous example. You have seen that by replacing each 0 in a binary string with a –1, you

get the corresponding bipolar string. If you keep all 1’s the same and replace each 0 with a –1, you will have a

formula for the above option. You can apply the following function to each bit in the string:

     f(x) = 2x – 1

NOTE:  When you give the binary bit x, you get the corresponding bipolar character f(x)

For inverse mapping, which turns a bipolar string into a binary string, you use the following function:

     f(x) =  (x + 1) / 2

C++ Neural Networks and Fuzzy Logic:Preface

A New Weight Matrix to Recall More Patterns

63

---

NOTE:  When you give the bipolar character x, you get the corresponding binary bit f(x)