0
$\begingroup$

I posted this on Software Engineering and was told that it might be better here and on AI. I'm currently reading Artificial Intelligence: A Guide to Intelligent Systems by Michael Negnevitsky (3rd edition) and came across this example of weight updates in a neural network. The equation for updating the weight in this network is: wi(p+1) = wi(p) + alpha * xi * e(p) where e(p) is the difference between the expected and calculated error, x is the input parameter, and alpha is the learning rate.The initial weights of the second epoch aren't different from the initial weights of the first. Aren't they supposed to be updated? What's happening between the final weights of an epoch and the initial weights of the next epoch? image 1 image

$\endgroup$
  • $\begingroup$ What exactly isn't different? $\endgroup$ – Tim Mar 7 at 16:43
0
$\begingroup$

Right, I think I see your confusion (I was confused too here for a minute). It seems that it is doing stochastic gradient descent (i.e, updating the weights after each input/observation).

So the way to read the table is row by row:

initial weights -> final weights (row 1)

then final weights of row 1 become initial weights of row 2 and so on so forth.

$\endgroup$
  • $\begingroup$ So only the weights on the very first line from the four lines of epoch 1 are being initialized with random numbers? $\endgroup$ – Omar Mar 7 at 18:07
  • $\begingroup$ Yes, that's correct! $\endgroup$ – Tom Mar 7 at 23:45

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.