0
$\begingroup$

I'm working on implementation of artificial neuron which be extended to neural net. I want do implementation by myself to fully understand how it works. I start with perceptron with threshold activation function:

enter image description here

and then questions come along

  • can implementaion be change to operate with more than 2 classes?
  • if I build network with few layers, first (and nexts) layers should return predicted class or value from first layer?
  • Is better artificial neuron to build neural net than perceptron?
$\endgroup$
0
$\begingroup$
  • The implementation can be changed for more than two classes. You can either use two neurons or you can use the following structure: $$ f(x)=\left\{\array{2 & \textrm{if} \;wx+b>0\\ 1 & \textrm{if} \; wx+b\leq 0 \; \textrm{and}\;ux+c>0\\ 0 & \textrm{otherwise} \\ }\right. $$ where $u$ is of the same size as $w$ and $c$ is a scalar.
  • Typically, only the output layers are used.
  • The described neuron is better in terms of interpretation but is difficult to be trained by gradient-based algorithms such as backpropagation.
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.