0
$\begingroup$

Is building a training data set, from unlabeled data, for a machine learning classifier considered as a scientific contribution?

$\endgroup$
2
  • $\begingroup$ What do you mean by "scientific contribution"? $\endgroup$
    – Tim
    May 21 '15 at 10:12
  • $\begingroup$ You should do and overview of related work regarding this topic. This will give you a quick overview of where you stand and if your solution is of scientific value. If I quote one of my professors: "If there is no related work regarding some topic, you are either trying to accomplish something brilliant or something stupid". $\endgroup$
    – alesc
    May 21 '15 at 11:20
2
$\begingroup$

You will very likely not be able to publish such a labeling as a standalone journal article - it is simply too routine and does not advance scientific knowledge as such.

If this labeling is done as part of a larger machine learning project, the contribution would be the larger project.

If, finally, there was one person in a team whose main contribution to a larger ML project was labeling unlabeled data, this may be sufficient for coauthorship; especially if the labeling took a significant amount of time and if this person also actively reviewed the manuscript etc. This would need to be clarified with the other authors. It seems to be on a par with the question when someone is a coauthor whose main contributions were writing software, or doing the statistics.


You may get better answers at Academia.SE. Consider editing your question to clarify what exactly you are asking and then potentially flagging it for migration. (Please do not simply cross-post.)

$\endgroup$

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