What should I do with the published paper if a section of paper is bad? [closed]

Question

What should I do if the paper is misleading (say deliberately done due to lack of money to buy hardware resources like GPU or lack of time)? What if a section of the paper has misleading information but, the rest of the paper is gold? What if it is just the Dunning Kruger or Impostor Syndrome making me think about retracting a highly cited paper (thinking I misled research community while in reality my paper would have helped research community a lot)? What if my paper has a plagiarised section but the rest of the sections are awesome?

Please note that I do not have any deep learning papers published. These are just interesting thoughts taken from A bad feeling about first published papers after somebody commented to make this another question.

Answer 1

If your paper contains plagiarized material, or material intended to deceive, then you should retract it.* If there is an honest mistake or something you overlooked, then consider writing a corrigendum/erratum or a follow-up article. Yes, this might include "we revisited problem X with new resources and show the previous results to be false/misleading". In each case the goal should be to correct the literature. This goes double for highly cited papers, which may cause a higher degree of damage. In future papers, make sure the claims are substantiated, even if that makes them weaker and the results less flashy.

*There might be some de minimis cases where this doesn't hold, but I think it works as a general principle.

Disclaimer: I work in physics, not a CS field. Not sure if that affects the approaches or ideals involved. I recently learned that it's apparently very difficult to publish errata in mathematics, for example. Still, correcting e.g. arXiv postings is a useful step.

Answer 2

Let's take your questions one at a time, starting with the preliminary one:

What if it is just the Dunning Kruger or Impostor Syndrome making me think about retracting a highly cited paper?

Discuss this with non-coauthors, and particularly people who are not biased in your favor and have not already been "sold" on the paper, to see what they think. This is a bit risky though, because they could theoretically abuse your thrust and bad-mouth you elsewhere before you've done anything yourself; and you're sort-of committing to accept their advice. Still, another look-over of the results by someone other than the authors, and a consultation, is probably the best way to neutralize the effect of impostor syndrome.

What if my paper has a plagiarized section but the rest of the sections are awesome?

@Anyon already addressed this case. However - if it's just a paragraph without references, you might be able to get away with a corrigendum and an apology. Coordinate this with your co-authors and advisor though.

What should I do if the paper is misleading (say deliberately done due to lack of money to buy hardware resources like GPU or lack of time)?

The cause of it being misleading is not so important; it's the effect that's the problem. So, first double-check that it's actually misleading (again, possibly using a third party). Now, if the mis-lead can be corrected with a clarifying addendum/erratum, and is not properly false, then consider submitting that. At any rate, you should probably contact the editor and consult them about this situation. (Of course, they might just tell you "Just retract it, I can't be bothered with this stuff" and you would need to retract IMHO.

In the case of a retraction, try adding some minor additional result and republish the paper, done right, without the inappropriate section's content.