Ever wondered how it’s possible you can talk to your phone and it’s actually giving relevant answers?
The answer is: Neural networks. A computer technology which works very similar to the human brain. And makes it possible for computer software to learn from example data (data sets).
How is this done?
It’s not easy. In the end it’s all about statistics and statistical analysis. One specific statistical analysis technique is the most important: Regression analysis.
Imagine the following dataset, consisting of x and y values:
Based on the data above, what would the y-value be when x = 18? The correct answer is: 36. Why?
Take a look at the data set and notice it’s simply doubling the data for x to get the y-value.
This is high school math and not very complicated.
We say y is a function of x.
In formula form: y=2x.
Now, let’s have a look at a real world example:
As you may notice, there’s no ‘fixed’ model. The data points are scattered around the real function. The function in this case is just a ‘best fit’ in the data set.
As you see, the function is a straight line. We therefore say it’s linear. This type of regression analysis is called ‘linear regression analysis’.
Below other ‘real world’ examples:
As you might see, the two examples above are non-linear. Both are plotted logarithmic. The number e, which is the base number for natural logarithms. By adjusting one or more of the axes to logarithmic is a common way of displaying logarithmic functions to display the functions in a nice way.
You may be able to think of even more complicated functions, such as quadratic, multi-dimensional or trigonometric. They all exist and are applied in machine learning.