1
$\begingroup$

Hi I was wondering if we can compare Noisy label problem to a semi supervised approach? Also are there any papers on learning with noisy labels? Any help is appreciated.

$\endgroup$
3
  • 1
    $\begingroup$ Define what is considered a noisy label. Instance duplicates with different labels? $\endgroup$
    – inzl
    Commented Aug 26, 2014 at 13:06
  • 1
    $\begingroup$ Duplicates with different labels that makes the class space overly fragmented $\endgroup$
    – user48918
    Commented Aug 26, 2014 at 13:31
  • 1
    $\begingroup$ Yes this is possible. I've done this in one of my papers. Preprint freely available here. $\endgroup$ Commented May 2, 2015 at 13:22

2 Answers 2

0
$\begingroup$

I would consider it as a semi-supervised learning problem. I would treat instances that have no contradicting duplicates as labelled instances, and instances with contradictory duplicates as unlabeled instances. Then it becomes a classical semi-supervised learning problem.

$\endgroup$
0
$\begingroup$

http://www.aaai.org/ocs/index.php/AAAI/AAAI15/paper/viewFile/9366/9948 Here is a paper start with initial noisy label data.

$\endgroup$

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.