CODEWITHSUNDEEP
matlabartificial-intelligenceneural-networkperceptron
Andrei
Andrei

Reputation: 37

Classifying Multiple Outputs using a Single Neuron

I am a beginner in AI so I am sorry if I ask a dumb question. What I am trying to do is to train only one perceptron to classify some input in 4 classes. I know that, normally, a perceptron is a binary classifier, but I am wondering if I can break the rule for this particular case which has a very easy classification rule. The data sample contains points given by cartesian coordinates. It is given the fact that x-coordinate will always be an integer between [1, 4] and the y-coordinate will take values between 0 and 1 with only 1 digit precision. ( [0:.1:1] Matlab expression). The classes depends only on X-coordinate. The class is just the X-coordinate. For example (1, 0.3) is in class 1, (3, 0.2) is in class 3. It is possible to train only 1 perceptron to learn this rule? Thanks in advance

Upvotes: 1

Views: 194

Answers (1)

krisdestruction
krisdestruction

Reputation: 1960

The answer is no, not with a standard binary bipolar threshold perceptron. Unless you have some crazy non-linear perceptron that has logic already built into the transfer function, you will require more neurons to approximate the logic. Keep in mind the more non-linear your transfer function is, the more complex your training will be.

Take the example of the XOR operator. Even a simple logic gate such as that will require 2 hidden neurons. Here's a reference of what I'm talking about for XOR.

For those questioning what the OP meant by binary perceptron, the OP meant this.

Binary Bipolar Threshold Perceptron

Upvotes: 2

Related Questions