4

I am an undergraduate student who is working on some research projects applying machine learning to medical problems at a small, primarily-undergraduate college.

In one of these projects, a Master’s student suggested a feature extraction approach and wrote some code based on it. Unfortunately, I couldn't reproduce the results. I tried to talk to him about it, but we couldn’t reach an agreement. I showed the professor leading the project that the results were not trustworthy, but she wants me to write a paper based on the method, where the other student and I would be listed jointly as first author.

In the meantime, I do not have any alternative results in hand. The problem we are working on is very difficult and has never been attempted before, and I am still developing the codebase for my own approach.

This conversation has been going on for a while, and my professor is getting impatient and wants to publish quickly. She says that some people within the department are pushing for her resignation (for reasons not pertinent to this question), and I think that may be the reason.

I am unsure of how to handle this situation because

  • I do not want to get the professor in deeper trouble

  • I want to remain a good member of our team

  • I want to do good work

I do not really have anyone who I can go to for advice without getting this professor in trouble, and I have no idea how to navigate this situation.

Why I couldn’t reproduce the results:

The model only performed well when trained and tested on all data, but under those circumstances, almost any model could, and many silly ideas can even outperform it. The technique didn’t really make sense within the domain, either, and I am not sure the student understands what it does.

Similar Questions

I have looked online for similar situations, like these ones, but I am still not sure what to do.

professor pushing me to get result when nothing is right!

Paper on project without any result - how to conclude

What to do with a result too small to be publishable, but interesting nonetheless?

How to proceed when the baseline (state-of-the-art) published results claim much better performance than I can reproduce?

Problem with undergraduate research supervisor, how to proceed?

Informing my supervisor of discrepancy in the results

My Professor is pushing me in a problem that isn't solvable for the deadline

Is it good to publish a research paper with many partially answered questions with sub-optimal results?

Is it morally correct to submit a paper to a conference with wrong results?

  • 4
    Be careful you do not taint your reputation with sloppy research. Let them sort out the authorship amongst themselves, training and testing on the same data is a severe breach of protocol unless you correct for the overfitting and they do not seem to do it. Avoid co-authoring that paper. – Captain Emacs Dec 31 '19 at 22:42
  • 1
    You can do that as a favour and decline being on the author list. Acknowledgements is ok for that. You have to know if you want to be co-responsible for the content (i.e. a co-author). – Captain Emacs Jan 1 at 8:57
  • 2
    Just beware that some professors know a lot. If not of their field, of what to do. It is a real possibility not to be ruled out. – Alchimista Jan 1 at 10:07
  • 2
    To the OP and to @CaptainEmacs, note that you aren't responsible for the reputation or future of the professor. Only she is. If you don't believe in the outcomes of your explorations, back away. – Buffy Jan 1 at 21:18
  • 1
    @Buffy That's precisely what I implied. Authorship means endorsement. – Captain Emacs Jan 1 at 22:01
1

You have two options. First option: decline to be the author of the paper, second option: write and submit the paper as first author. In the second case, the paper will go under review and the reviewers and the editors will decide if it is a good publishable result. Personally I would choose the second option in a high-medium ranking journal. In this way, if the paper will be rejected, it will not be your fault and you will not enter in contrast with your professor.

| improve this answer | |
0

You can convince your professor with more tangible datas. However, you have to master the statistical calculating methods for this and have to find the quality articles which reflect well the false one.

| improve this answer | |
0

It unfortunately sometimes happens that a professor or other PI may see a result from a student that they want to be true, but then a more junior student fails to reproduce the result, and the professor trusts the more senior student.

If you have time and it is not too stressful for you, writing the paper can be good practice, and may help you communicate more effectively with your coauthors. It is always possible that you are the one who is confused, and you will learn from the writing process. It may also be that once your coauthors see it in writing, they will understand better what you are trying to say.

I agree with @Gab that the reviewers can and must ultimately decide whether the work is publishable, assuming:

  • there is blind review (they won't know the authors) and/or your coauthors or institution are not particularly famous. Note that more fame is more power and entails more responsibility. Reviewers may give some people the benefit of the doubt. You can say the world shouldn't be like this but a) reviewers are human, and b) actually, you get more certainty from knowledge from multiple sources. So there is a matter of professional responsibility too.
  • you don't allow anything with your name on it that you do not believe to be true, or that you believe to be deliberately deceptive.

As long as everything in the paper is factual, the significance of the work might be better judged by other authorities.

If you do think someone is trying to deceive someone else, then you may have an obligation to talk to the supervisor of whoever the lying party is.

| improve this answer | |
0

If you can’t reproduce the results at all, I’d back off. Don’t put your name on something you don’t trust.

But feature extraction techniques, SEM, CFA etc are well known to be tenuous and very sensitive to the data. Any informed reader will (should?) already appreciate that.

| improve this answer | |

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.