There is a quiet tension at the heart of this JMLR submission question, and it is not about review timelines or tenure requirements. It is about what happens when a field's most prestigious venue starts to feel like a foreign country to the very people who should feel most at home there. The PhD student is not asking for a simple logistics breakdown. They are asking whether the map they have been given still points to the right territory. Their supervisor, a statistician with a stellar record in finance and actuarial journals, is pushing for a venue that the student's own computer science circle treats with suspicion, or worse, with the weary resignation of someone who has watched peers get burned after years of silence.
That gap in perception is the real story here. The student notes that in statistics, a three-year review process usually ends in acceptance, while in machine learning, a two-year wait can end in a rejection that leaves a paper obsolete before it can be resubmitted. This is not a minor procedural difference. It is a fundamental mismatch in how knowledge is validated and rewarded. The supervisor's instinct to favor journals over conferences is reasonable given his own field's norms, but it reveals a deeper problem: the metrics that once signaled rigor and prestige are no longer aligned with the pace or the practical realities of modern AI research. We have seen this tension play out in related discussions, such as Explore the Forrester Function: Beyond Mathematics, a Tool for Machine Learning, where mathematical elegance meets applied pressure, and in Unlock LLM Training: A Practical Guide to Distributed Algorithms, where the field's focus on scalable results often overshadows slower, more deliberate paths to publication.
The student's specific concerns about reviewer quality are telling. They have experienced conference reviewers who copy-paste from language models and demand impossible proofs, while their stats journal reviewers at least offer mathematically sound feedback. This is not a complaint about rigor; it is a complaint about accountability. In a world where Exploring Paragraph Structure: How LLMs Navigate Token Space shows us that even the structure of our outputs is being reshaped by AI, it is no surprise that reviewers are using the same tools carelessly. But the student is asking a fair question: if the reviewers do not understand the math, and the editors cannot judge it, what exactly is being evaluated? The answer, in too many cases, is inertia. A bad review in a conference can be dismissed; a two-year silence from a journal is a different kind of verdict, one that offers no path to appeal.
Here is the takeaway we would offer to anyone in this position: do not confuse prestige with fit. JMLR is a legitimate and respected venue, but it is not the only venue, and it is certainly not the right venue for every paper with a proof and an application. The supervisor's tenure committee may value journals, but the student's career depends on timeliness and on feedback that actually improves the work. If the supervisor is unwilling to engage with the computer science literature, that is not a problem with the venue, it is a problem with the collaboration. The student should push for a dual strategy: submit to a strong conference with a shorter cycle, and simultaneously prepare a journal version for a statistics or applied mathematics outlet. That way, the work is not held hostage by a single review process. And if the supervisor insists on JMLR, ask him to read three recent papers from it and explain how our work fits the editorial vision. If he cannot do that, he is not ready to navigate that world, and the paper will pay the price. The question is not whether JMLR rejects fast enough. It is whether you are willing to let someone else's timeline decide your relevance.