Practical Neural Network Recipies in C++

18

Chapter

2

subjective threshold. If, on the other hand, we cannot be positive that

it is a strictly binary decision, we must use two output neurons and

approach it as a multiple-class problem. This is discussed in the next

section. Then, the activations of the two output neurons can help us

decide whether a difficult case is a member of one class but is also

similar

to

the other (having both neurons highly activated), or whether

it is atypical of both classes (having both neurons nearly off). Unfortu-

nately, we still have the potential problem of both neurons having

intermediate activation. In this case, we are back to the situation of

inadequat ...