1
$\begingroup$

I'm trying to understand some basic neural network stuff and I'm having problems with one thing: Let's assume that we have to classify two classes: red and green dots. To do that I create three lines on a diagram:
diagram

It's easy to calculate functions responsible for them from diagram:
1: y = 1/3 * x + 2
2: y = 6/7 * x + 3/7
3: y = 1/3 * x - 3

Next step is to design neural network and assign weights:
enter image description here
And thats the part I'm not sure about. Input layer weights should be taken from equations?

  • 0
    Why do you have three inputs? Surely your inputs are x and y – 2017-01-16
  • 0
    Are you sure that the third (lowest) line on your graph is needed? Are all dots below the line supposed to be classified as green? – 2017-01-16
  • 0
    I assume that I should have three inputs as I have three different equations, not only one linear function. – 2017-01-16
  • 0
    Sorry @Hugh for misleading , I edited my question with proper image with green dots under third line also. – 2017-01-16
  • 1
    Each neuron is an example of a linear equation which is of the form $\mathbb{w}^T\mathbf{x}$ where by when expanded, results in $w_0 + w_1x$, therefore your weights are easily shown from the three equations - $w_0$ is your bias (shift). Note, that you can use other functions as your basis function. – 2017-01-16
  • 0
    @Chinny84 , so instead of usng y = ax+b to determine the weights I should calculate : $y = w_{n}x + 1$ for n = 1,2,3? Is this $w_{0} $ can be different for each of the neurons in different case? – 2017-01-16
  • 0
    yes each bias will be different as each coefficient of your input can also be different . Each neurone is a model in itself and is free to fit to whatever it deems fit to. It is better to think of your $\mathbf{x} = [1,x]$ rather than how you have shown. – 2017-01-16

0 Answers 0