8

I have been examining a novel machine-learning framework for a seminar at uni. After some people asked me whether I would publish the results in a paper so they could see them as well, I thought I might as well put some more effort into it and make a publication out of it.

However:

  1. I am not sure whether this adds any research value since I did not really do something new, I just summarized the key concepts of the framework and took example use-cases to compare their implementation with and without the framework (implementations without the comparison have been done in blog posts).
  2. I have recently been told that publications in a non-high-ranking journal can actually hurt your career. I do not want this to negatively affect my PhD application later this year.

My question is thus: Is it worth publishing a paper that is only an examination of a framework and a comparison with not using it?

Some technical background: the framework in question is Google's TensorFlow and I am comparing plain Python implementations of machine learning algorithms with the implementation utilizing the framework. I wanted to also briefly summarize the framework and compare it with other machine learning frameworks already out there - this is only half a page regarding performance, capability etc though, no code is compared here. Ultimately, the paper aims at people currently using Python who are not sure whether they should switch.

5

I think that there are two questions in here that it may be useful to separate out:

  1. Should you spend the time to prepare a well-written analysis that you make publicly available to others?
  2. Should you make your analysis available in the form of a peer-reviewed paper?

For the first, it seems that you've done much of the analytical work already, and that there are people who will appreciate it. I thus see no reason not to produce a good write-up and publish it.

Whether you should publish it as a peer-reviewed paper, on the other hand, is not so clear to me. I don't think that you need to worry about being looked down upon for publishing such a paper: that concern is more about bad work, not work with a clear, intentionally limited scope. I just don't think there's much to gain from going through a formal peer review process here unless your work involved more experimentation and surprises than it sounds like it did.

Instead, I would recommend posting your analysis as a technical memo (many institutions have a mechanism of this sort), a whitepaper, an arXiv pre-print, or other such semi-formal document. A semi-formal publication of this sort can place your work in a permanent archive with a DOI and clear means of citing if people find it useful, yet also is fast, easy, and need not be a significant distraction from the main thrust of your research.

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.