My task is to learn a simple neural network with input size 1, hidden layer size 8, and output size 1 (a function $f = {0->1,1->0,2->0,3->1}$, but can learn the network to get satisfying result.

I am using an activation function, back propagation, and initial values in range (-1,1).

However, I can not get better results in supremum error norm than 0.5, never achieved under 0.5.

What did I do wrong? That is only 4 examples to learn, why is so hard to learn that natwork? 2 years ago I tried to learn XOR in similar way and also had same problem.


Your Answer

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

Browse other questions tagged or ask your own question.