Journal For Machine Learning Top 10

What You Actually Need to Know Before Submitting Your ML Paper

I have spent the last several years watching people waste months on the wrong venue for their work. The problem is not that there are no good journals. The problem is that most researchers do not understand how different they actually are, and they submit to the first one that sounds prestigious enough on paper. There is no single ranked list that matters universally. But if you compile the Journal For Machine Learning Top 10 based on actual publication velocity, peer review quality, and citation impact over the past five years, you get a specific set of outlets that the field actually respects. Everything else is noise.

Journal For Machine Learning Top 10

Here is the breakdown. I am not ranking these 1 through 10 in strict order because the differences between positions 4 through 7 are marginal and depend heavily on your subfield. What matters is knowing which ones fit your work. This is the open access standard. It is free to publish in and free to read. The review process is genuinely double blind. I submitted a paper on reinforcement learning calibration here once and the reviewers spent eight weeks going through my ablation studies line by line. The feedback was harsh but useful. The acceptance rate sits around 20 percent. If your paper is long and methodology heavy, this is the right place. A solid traditional journal with a slower review cycle. Papers here tend to be more theoretical. I have seen reviewers demand full mathematical proofs for claims that JMLR would accept with empirical validation. The submission to first decision usually takes four to six months. Good for work that needs time to breathe rather than racing toward a conference deadline.

The heavy hitter for computer vision and pattern recognition adjacent work. Review times are long, sometimes eight months. The bar is high and the rejection rate is steep. One thing people do not tell you: TPAMI reviewers heavily weight novelty of formulation over novelty of application. A clever application of an existing method will get desk rejected. A boring application of a new theoretical framework has a chance. Similar to JMLR in open access style but with a broader AI scope. The editorial board is large and decisions can feel inconsistent. I once had a paper rejected here for being too narrow, then accepted at JMLR three months later with minimal revisions. The review quality here varies significantly depending on which editors you get assigned. One of the older journals in the field. It leans toward symbolic AI, reasoning, and knowledge representation. If your work is purely statistical or deep learning focused, this is not the right venue. The editorial team prefers papers that make claims about general AI progress rather than incremental improvements on benchmark datasets.

Mit Press journal. Good for work that bridges neuroscience and machine learning. The mathematical bar is high. I have seen perfectly valid engineering papers rejected because the authors did not justify their architectural choices from a computational principles perspective. Read a few recent issues before submitting to understand what they actually want. Broader scope than TPAMI. Accepts work on neural architectures, learning theory, and applied systems. The review process is faster than TPAMI, usually three to four months. This is a common landing spot for solid applied deep learning work that does not need the prestige of the top vision venues. Not a research journal in the traditional sense. It publishes survey and review papers only. If you are writing a comprehensive literature review on any ML subfield, this is the target. The bar for acceptance is extremely high. They typically only accept survey proposals that are invited or come from established researchers in the area. Do not submit a survey on a topic you entered two years ago.

This is a niche option. Relevant if your work has a strong statistical foundations component. Not many ML researchers target this journal specifically, which means less competition but also less visibility within the mainstream ML community. Focuses on multi modal and multi source learning systems. If your paper deals with fusion of heterogeneous data types, this is the venue. The impact factor has risen steadily. Reviewers tend to be more applied than theoretical. I want to address a specific problem I ran into that most people never expect. I was preparing a submission to JMLR and included a supplementary material file containing my training logs and hyperparameter sweeps. The editorial system rejected the submission because the supplementary file exceeded their file size limit by approximately 40 megabytes. The error message was vague. I spent two days trying to compress the files without losing readability of the tables. The workaround was straightforward but not obvious: I moved the raw logs to a public GitHub repository and referenced it in the paper instead of attaching them. JMLR accepts external links for supplementary material as long as the link is permanent. I wish the submission guidelines mentioned this explicitly.

Here is a counter intuitive point that beginners miss. Citation impact and journal quality are not the same thing. A paper published in a lower ranked journal that solves a real problem will be cited more than a paper in a top journal that makes a marginal contribution. I have seen this repeatedly. Do not confuse venue prestige with research value. The citation metrics on these journals are skewed by review articles and survey papers that naturally accumulate more citations over time. Another thing nobody talks about openly. Some of these journals have different implicit preferences depending on the geographic region of the editorial board. TPAMI and TNNLS tend to favor work from labs with large engineering teams. JAIR and JMLR are more open to work from smaller groups. This is not a rule, just an observed pattern across hundreds of submissions I have tracked. The biggest mistake I see is people treating journal submission like a conference submission with a longer review window. It is not. Journals expect you to have done more extensive experiments, more thorough ablations, and more careful related work positioning than you would for a conference. A conference paper with seven experiments might get a journal review expecting twelve. Prepare accordingly or expect a rough process.

If you are early in your career, start with Machine Learning (Springer) or TNNLS. They are more forgiving of incomplete theoretical frameworks. Move upward to JMLR and TPAMI once you have a clearer sense of what kind of rigor your specific subfield demands. The Journal For Machine Learning Top 10 is not a ladder you climb linearly. It is a set of tools, each designed for a different type of work. Match the tool to the job and you will save yourself a lot of rejected submissions.