Non-linear classification example
XOR / XNOR function
XOR and XNOR (not XOR) problem can classifify by neural network with complex function.
(source: https://www.coursera.org/learn/machine-learning)
AND function
Recall sigmoid function $g(z)$, there are some values to know:
$$g(4.6)\approx{0.99}$$
$$g(-4.6)\approx{0.01}$$
OR function
(source: https://www.coursera.org/learn/machine-learning)
According to the on-line class quiz above, we know that
$$g(x_i)=g(-10+20x_1+20x_2)\tag{1}$$
$x_1$ | $x_2$ | $g(x_i)$ |
---|---|---|
0 | 0 | $g(-10)\approx0$ |
1 | 0 | $g(10)\approx1$ |
0 | 1 | $g(10)\approx1$ |
1 | 1 | $g(30)\approx1$ |
Observe the result in the right column of the table, it is only false at $x_1=0$ and $x_2=0$, so $g(x_i)$ here is an OR function.
General Idea
Negation (NOT)
If we need a not function, then put a large negative coefficient of $x_i,i\ne0$.
For example, if we want a (NOT $x_1$) AND (NOT $x_2$)(i.e. if and only if $x_1=x_2=0$ the result will be TRUE) function, the following function will be feasible.
$$g(x)=10-20x_1-20x_2$$