In neural network, we should use a bias, for example:
y = w•x + b
where b is the bias.
Why we need use a bias in neural network? In this tutorial, we will answer this question.
Look at image below:
There are two kinds of points in 2 dimension space. In order to separate them, we can use a linear function.
y = w•x + b
Bias b must be need, otherwise, we will not seperate these two kinds of points correctly.
However, if we set b = 0, which mean we do not use bias, what problem will occur?
It means y = w•x
In order to sperate these two kinds of point correctly, we should move all points.
Which means
y’ = y – b
However, it is very hard or impossible to move sample data to a specific point to make bias = 0. Especially there are some noise data in your training data. So in order to increase the accuracy of neural network, we should use a bias.