P5_Perceptron
Description
Simple pattern recognition using a perceptron and supervised learning.
Supervised learning algorithm
-
Give inputs to the perceptron for which there are known answers.
-
Perceptron guess an answer.
-
Compute the error.
-
Adjust all the weights according to the error.
-
Return to 1. and repeat.
Perceptron
In the modern sense, the perceptron is an algorithm for learning a binary classifier: a function that maps its input x (a real-valued vector) to an output value f(x) (a single binary value):
f(x) = 1 if w*x + b > 0 | = 0
Basically, a perceptron recieves an input and computes a guess (output).
-
Point living over the line = Output of +1
-
Point living under the line = Output of -1
Error correction
Error is calculated with this formula:
Error = desired output - guessed output
| Desired | Guessed | Error |
|---|---|---|
| -1 | -1 | 0 |
| -1 | +1 | -2 |
| +1 | -1 | +2 |
| +1 | +1 | 0 |
So the possible error values are: -2, 0, +2
Error correction is done by using the error value in the weight formula:
delta weight = error * input
new weight = weight + error * input * learning rate
Previews
Setup
-
Open project with the Webstorm ide and open the index.html file.
-
On the top right corner select Chrome browser to open the project in Google Chrome.