Thoughts and experience on ML journals [D]

A machine learning researcher's candid post about considering a shift from conferences to journals has gone viral, tapping into widespread frustration with the peer-review process at major AI venues. The author cites "a few bad experiences with ML conferences" as the catalyst, noting a lack of personal experience with journals and a tendency not to read journal papers. This admission underscores a core tension: ML is a famously conference-centric field, where premier venues like NeurIPS, ICML, and ICLR drive rapid dissemination and career advancement, yet their review systems are often criticized for inconsistency and opacity.

The post specifically rules out the Journal of Machine Learning Research (JMLR) due to "extremely long waiting times" and a mismatch with the author's typically shorter papers. Instead, the focus turns to alternatives like Transactions on Machine Learning Research (TMLR), Neurocomputing, Neural Networks, and the journal Machine Learning. The author seeks practical insights into their selectivity and overall quality, questioning what the official "Q1" journal ranking truly means in the fast-paced, reputation-driven ML ecosystem. This sparks a broader discussion on whether established journals can offer a more rigorous, thoughtful, and stable publishing path compared to the high-stakes, sometimes chaotic conference cycle.

The resulting discussion thread has become a valuable crowdsourced resource, with researchers sharing detailed experiences on review timelines, editorial standards, and the perceived prestige of various journals versus conferences. The conversation reveals a community grappling with the trade-offs between speed and thoroughness, impact and stability, in how foundational AI research is communicated and evaluated.