-1
$\begingroup$

For back propagation for a multi layered perception, are all the gradients for each weight calculated in each layer first through back propagation, and then all of the gradients for each weight updated with respect to its calculated gradient and the learning rate?

Is this how back propagation works for a multi layered perception?

Thanks and appreciate the help!

$\endgroup$
-1
$\begingroup$

For every neural network architecture out there: during backpropagation, all weights get processed and thus possibly modified.

So the 'wrongness' of every neuron in a neural network gets calculated (aka error). With this 'wrongness', we can see how much every input connection was responsible for that 'wrongness'. The weights get updated so they are less wrong in the next forward propagation.

$\endgroup$
  • $\begingroup$ This does not provide an answer to the question. To critique or request clarification from an author, leave a comment below their post. - From Review $\endgroup$ – Antoine Jun 16 '17 at 11:01
  • $\begingroup$ @Antoine question title: Are all weights updated during backpropagation for a MLP. My answer: all weights get processed. How is that not an answer to the question? $\endgroup$ – Thomas W Jun 16 '17 at 11:50
  • 1
    $\begingroup$ It seems a bit brief to help the OP with his/her understanding since it does not convey much more than a simple "Yes". $\endgroup$ – mdewey Jun 16 '17 at 11:53
  • $\begingroup$ Thanks for the reply. So all weights are potentially updated during back propagation? So for back propagation, it starts with the error and back propagates to update each of the previous layer until it gets to the input layer? Does anyone have a good mathematical paper or link to an example of back propagation for an MLP? Thanks and good job! $\endgroup$ – Bo Peng Jun 16 '17 at 15:07

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.