For activation function in neural network, I have used the function $$\frac{1}{2}(\sin x +1) \enspace .$$ But this will give me the value of either 1 or 0, allowing me to classify only 2 classes. What if I want to classify more than 2 clusters of data, then what activation function should I use?
$\begingroup$
$\endgroup$
$\begingroup$
$\endgroup$
To divide the data into K classes you can use K single layer perceptrons. Perceptron k is trained to output 1 if a training data belongs to that one. When using the perceptrons, calculate the output of the K perceptrons and assign the test data to the class with largest perceptron output.
You can save 1 perceptron by considering the fact that an input should belong to at least one class. For more information look up: http://en.wikipedia.org/wiki/Multinomial_logistic_regression
