1
$\begingroup$

Suppose I use an auto encoder for feature learning on a certain dataset. How can I use the learnt features e.g. for a classification task?

Should I feed a SVM with the reconstructed output or with the weights out of the hidden layer?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
0
$\begingroup$

It's neither the reconstructed output (since we wouldn't have dimensionality reduction here and we want to produce compact features) nor the weights themselves (because they are somewhat independent from the input and first need to be applied to the input via cross-products).

It's as simple as it gets and we just need to encode the input data and use this hidden representation (reduced in dimensionality) in our SVM.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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