My First Research Paper Got Rejected 4 Times. The 5th Submission Accepted It.

My first paper was about an adaptive sampling strategy for imbalanced classification datasets. I thought it was good. Four peer review panels across four conferences thought it had merit but also had significant problems that I needed to address before it was publishable. They were all right. This is the story of how being wrong four times in a row produced a piece of work I am actually proud of.

The first rejection came from a workshop at a top ML conference. The reviewer comments were two sentences long. I spent three days trying to parse them for actionable feedback before my supervisor told me that two-sentence reviews are usually from a reviewer who read the abstract and nothing else, and that I should move on. I moved on.

The Second and Third Rejections

The second rejection, from a regional conference, came with three detailed reviews. One reviewer thought the experimental setup was sound. Another thought the baseline comparisons were insufficient. A third asked why we had not compared against a specific method published the previous year that I had genuinely not been aware of. All three were correct in their own ways. I spent six weeks after this rejection rewriting the experimental section, adding two baselines, and reading everything that had been published in the space since 2021.

The third rejection was the hardest. We had addressed all the previous feedback. We submitted to a better venue. Two reviewers praised the work. One gave it a score that made acceptance impossible and whose main argument was that the problem was not important enough. This is a philosophical rejection, not a technical one, and it is the kind that requires the most mental resilience because you cannot fix it with more experiments. You can only decide whether you agree with the reviewer's framing of importance or whether you think they are wrong, and proceed accordingly.

What the Rejections Did to the Paper

The version of the paper that was eventually accepted at a solid workshop at ICLR was almost unrecognisable from the original submission. The core idea was the same. The clarity of the writing had improved enormously because I had been forced, four times, to explain to hostile readers why this mattered and how the work actually performed. The related work section was thorough in a way it would never have been if I had gotten lucky on the first submission. The limitations section was honest in a way that first-time authors rarely manage.

I tell my junior colleagues who are new to research: a first paper rejection is information, not failure. A fourth paper rejection is a lot of information. Be grateful for every reviewer who was specific, even when the specific thing they said hurt. The ones who are vague are not teaching you anything. The ones who are precise are doing you a service even when it does not feel like it.