2

I am currently working on my Master's thesis in machine learning and recently encountered a research paper that addresses a similar problem to my thesis topic. Both approaches aim to enhance performance using label information from the data, albeit in different ways.

The methodology presented in the paper involves additional steps, which my approach circumvents by optimizing a different objective, potentially making it simpler and more direct. Additionally, the existing method has some drawbacks, such as generating unused data and potential overfitting issues—problems that my method avoids.

Given these differences and the shared goal of our work, I am worried about how to address this paper in my thesis.

Would it be problem to discuss this paper in the related work section of my thesis? Should I emphasize the differences in methodology and the potential downsides of the approach found in the paper? Or does this not make any sense and I have to find a new topic? I appreciate your advise.

Improve this question
1
  • 3
    Though I don't publish in your field, I would take "aim to enhance performance using label information from the data" as descriptive of nearly every paper published in your field, so if you've found only one that is similar either you're not looking very hard/with the correct search terms or your approach is quite novel to only be similar to one other paper.
    – Bryan Krause ♦
    Commented May 10 at 12:57

1 Answer 1

Reset to default
9

This happens a lot when you work in a popular area. It has happened to me on several occasions in math and stats.

Would it be problem to discuss this paper in the related work section of my thesis?

It's not a problem. This is exactly the right thing to do.

Should I emphasize the differences in methodology and the potential downsides of the approach found in the paper?

Yes, but try not to be too negative about the other paper. For sure you should emphasize the ways in which yours can be viewed as an improvement, including having less risk of overfitting.

Or does this not make any sense and I have to find a new topic?

No, don't do that unless your advisor says so. Even if you found another topic, as you got deeply into it, you'd probably find related work to that topic, too. This is the nature of research in machine learning. A whole lot of research as been done! A bit of overlap with existing research is not a big deal, especially when you can point to the differences and can argue that your approach can be viewed as an improvement of existing approaches in some way.

Improve this answer
4
  • Thank you very much for your advice. In this area, several studies try to improve performance in related ways, but frame it differently. How much overlap might be tolerable. The goal is, as I said, similar and we use the same additional information, but the execution is carried out quite differently.
    – Simon
    Commented May 10 at 12:36
  • 1
    I have seen papers with quite a lot of overlap still get published. I wouldn't stress too much about this if I were you. If you are stressing you can talk to your advisor. They can also smooth the publication process for you once you get to that point.
    – David White
    Commented May 10 at 12:59
  • But if it is only a master thesis, should it be enough?
    – Simon
    Commented May 10 at 13:08
  • Hello it is me again. I implemented the related algorithm and found out that it performs much worse. Should I include the discussion of the approach in the main part of the thesis and also give the results ?
    – Simon
    Commented May 11 at 8:15

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .