In most sciences, prestige comes from publishing in journals. In AI and machine learning, it comes mainly from conferences. Researchers race to submit papers to a handful of top venues, and the results — accepted papers, spotlight talks, best paper awards — shape careers and research agendas. Here is a guide to the most important ones.

Why conferences matter in AI

AI moves fast, and conferences offer quicker review cycles than journals: typically a few months from submission to decision. Top conferences are highly selective, and accepted papers are published in proceedings. Many papers also appear first as preprints on arXiv, so the community often discusses work before formal acceptance.

The big general machine learning conferences

NeurIPS

The Conference on Neural Information Processing Systems, held each December, is the largest and most prominent machine learning conference. It started in 1987 and has grown to receive many thousands of submissions and attract well over ten thousand attendees. It covers the full range of machine learning, from theory to applications, and hosts many workshops and a datasets and benchmarks track. The conference changed its acronym from “NIPS” to “NeurIPS” in 2018.

ICML

The International Conference on Machine Learning, held each summer, is one of the oldest machine learning conferences, with a strong focus on methods and theory.

ICLR

The International Conference on Learning Representations, founded in 2013 by Yoshua Bengio and Yann LeCun, focuses on deep learning. It pioneered open peer review on the OpenReview platform, where reviews and author responses are public.

Domain conferences

  • CVPR (Computer Vision and Pattern Recognition), plus ICCV and ECCV, are the leading computer vision conferences. CVPR is among the most cited publication venues in all of science.
  • ACL, EMNLP and NAACL are the main natural language processing conferences, organized by the Association for Computational Linguistics and related bodies.
  • AAAI and IJCAI cover AI broadly, including symbolic methods, planning and knowledge representation.
  • CoRL and ICRA focus on robot learning and robotics.
  • FAccT (Fairness, Accountability, and Transparency) addresses societal impacts of algorithmic systems.

How peer review works

  1. Authors submit anonymized papers by a deadline.
  2. Several reviewers — volunteers from the community — assess novelty, correctness, clarity and significance.
  3. Authors may respond to reviews in a rebuttal period.
  4. Area chairs make recommendations; program chairs decide.
  5. Accepted papers are presented as posters, with a selection given oral or spotlight presentations.

The massive growth in submissions has strained the system, leading to concerns about review quality, randomness in decisions and reviewers using AI tools inappropriately. Conferences have introduced policies on AI-assisted reviewing and writing.

Industry presence

Major technology companies sponsor conferences, recruit at them and publish many accepted papers. Big product demos and parties have become part of the scene, especially at NeurIPS.

Industry and developer events

Separate from research conferences, technology companies hold developer conferences where they announce AI products and models, and many industry conferences focus on business adoption of AI.

Attending virtually

Since the COVID-19 pandemic, most major conferences publish recorded talks and posters online, and some offer virtual registration — making them more accessible to people who cannot travel.

Reading conference papers efficiently

With thousands of accepted papers, nobody reads them all. Start with oral and spotlight papers in your area, best paper awards and workshop overviews, and use paper search tools and community summaries to find relevant work. Read the abstract, figures and conclusion first, then decide whether a full read is worthwhile.

What to watch next

  • Changes in peer review to cope with growing submissions.
  • Policies on AI-assisted writing and reviewing.
  • Conference locations and visa accessibility for global researchers.
  • Growth of datasets, benchmarks and position paper tracks.

Key takeaways

  • Conferences, not journals, are the main publication venues in AI.
  • NeurIPS, ICML and ICLR are the leading general machine learning conferences.
  • CVPR leads computer vision; ACL and EMNLP lead NLP.
  • ICLR pioneered open peer review on OpenReview.
  • Rapid growth strains peer review quality and consistency.

Frequently asked questions

Can anyone attend NeurIPS?

Yes, registration is open, though tickets can sell out. Many talks are also available online.

Why are papers on arXiv before conferences?

Researchers share preprints to establish priority and get feedback quickly, given the fast pace of the field.

What is a workshop paper?

A shorter or more preliminary paper presented at a workshop attached to a main conference, usually with lighter review.