I am writing a machine learning manuscript for publication.

In a figure I had two learning curves for parameter 1 (P1) and parameter 2 (P2). Now, the reviewer asked to incorporate parameter 3 (P3) and show the learning curve for that.

The assumption is that the curve for P1 is better than P2 (P1>P2) and the curve with P2 is better than with P3 (P2>P3).

However, after training with parameter 3 I found a curve that shows (P3>P2). Probably P2 curve was stuck at a local minima. Since, my original P1 curve was okay, I was not bothered with the P2 curve that much when I submitted the manuscript.

Since, it is not expected, I retrained all three curves and after retraining I found the desired curves (P1>P2 and P2>P3).

Now, can I put this newer curves in the figure and submit the manuscript? Or I have to keep the original P1 and P2 curves intact?

  • 2
    How do you know your other curves aren’t stuck in local minima, too? Is it feasible to train multiple times to be sure your expected results are common or anomalies (I know training can be time consuming)? It’s ok to amend and it’s the right thing to do but be prepared for further scrutiny. – Pam May 4 '18 at 21:27
  1. I'd do it. But I'd be also be clear to all people involved, that I did it. See next item.
  2. That's exactly why we are often asked to highlight the changes to the manuscript after revision and why the "Response to reviewers" document exists. You basically write an accompanying file where you state "Reviewer 7 suggested to visualise the learning curve for P3. The curve is better than one for P2, making the old version of Figure 254 obsolete. The new version of Fig. 254 shows the learning curve for P3 as (f)."
  3. I would thank "the anonymous reviewers" for "improving our parameter choice" (or something like that) in the acknowledgement section.

You are allowed to improve a manuscript during review in areas other than what the reviewers specified!

Just make sure that you call attention to the changes and explain why you made them somewhere in your response to the reviewers.

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.