Perceptrons have a binary input of 1 and 0, but now that we’ve added weights, the output will almost always be greater than 0. What we’d really like is a sliding scale between one and zero, the closer we are to 1, the more confident the network in voting buying the fruits, the closer it is to 0, we need to change other places for buying. Therefore we’ll use a sigmoid function (sigmoid neurons) to squeeze that number to something between 0 and 1. Also, you should have some threshold to determine whether your neuron’s above a certain level, then you should buy fruits.
You want to see what number will make your neuron fire or stay silent, what we should do is creating a bias. The BIAS is one of the neural network’s HYPER-PARAMETERS which you might spend a lot of time tuning, almost like you’re adding a little weight to the scale. As the previous fruits perceptron, 3 for X1, 2 for X2, 6 for X3, I’ll also add a 5, just before I squeeze down my output. This bias will give me one more dial to fine-tune my fruit network to classify the supermarket as good or bad place for buying fruits.
