The International Conference on Machine Learning (ICML) is an annual academic conference dedicated to machine learning, first held in 1980 and organized by the International Machine Learning Society (IMLS). It is recognized as the oldest and, together with NeurIPS and ICLR, one of the three primary conferences of highest impact and reputation in machine learning and artificial intelligence research. The conference typically takes place in July, with paper submissions due at the end of January, though precise dates vary by year.
ICML serves as a central venue for presenting advances in the theory and application of machine learning, attracting researchers from academia and industry worldwide. Its proceedings are published open-access, and the conference has played a foundational role in the development of modern artificial intelligence technologies.
History
ICML originated as the International Workshop on Machine Learning, first convened at Carnegie Mellon University in Pittsburgh in July 1980. The workshop was organized by Jaime Carbonell, Ryszard S. Michalski, and Tom M. Mitchell, and it ran for over a decade before transitioning to a full conference series under the current name in 1993. Since the 2010s, its proceedings have been published in the open-access Proceedings of Machine Learning Research (PMLR), which evolved from the Journal of Machine Learning Research's Workshop and Conference Proceedings and was renamed PMLR in 2015.
Over the decades, ICML has published numerous influential papers that underpin contemporary machine learning. Notable examples include Lawrence Berkeley National Laboratory's work on K-means clustering via principal component analysis (2004), Google's introduction of batch normalization, Google Brain's EfficientNet (2019), OpenAI's CLIP (2021), and UC Berkeley's Soft Actor-Critic (2018). These contributions have shaped areas such as deep learning, neural networks, and generative AI.
The scale of the conference has grown dramatically. Submissions rose from 1,037 papers (270 accepted) in 2015 to 12,107 papers (3,260 accepted) in 2025. In 2026, submissions reached a record 24,371, more than doubling the prior year. Acceptance rates have fluctuated between 21% (2020) and 30% (2024). Accepted papers are presented as posters, with a subset selected for oral presentations or designated as "spotlight papers" by a committee. In 2026, 2.2% of submissions received spotlight status and 0.7% were chosen for oral presentation. Additionally, a jury annually honors the best contribution with the Outstanding Paper Award, formerly known as the Best Paper Award.
Scope and Focus
ICML's scope encompasses the full breadth of machine learning, with particular emphasis on theoretical analysis, algorithmic innovation, and statistical learning. Compared to NeurIPS and ICLR, ICML traditionally features more content on statistical learning theory, reinforcement learning, robotics, and optimization theory. This focus attracts researchers working on foundational questions as well as applied systems.
The conference's annual proceedings are published in PMLR, ensuring broad accessibility to the research community. This open-access model has contributed to ICML's role as a key resource for practitioners and academics alike.
Industry Engagement
ICML attracts substantial sponsorship from technology companies seeking access to cutting-edge research and emerging talent. Major firms such as Google, Microsoft, Amazon, Meta, and Apple regularly sponsor the event, publish their research there, and recruit researchers on-site. Financial institutions, including Citadel Securities, Jane Street Capital, and D.E. Shaw, also maintain a notable presence, reflecting the growing importance of machine learning in quantitative finance.
This industry involvement helps bridge academic research and commercial application, with many breakthroughs presented at ICML later finding their way into products and services. The conference also serves as a networking hub, fostering collaborations between university labs and corporate research divisions.
Impact and Recognition
ICML is widely regarded as a top-tier venue in machine learning, with a rigorous peer-review process and a strong track record of publishing seminal work. Its papers frequently introduce methods that become standard tools in the field, from optimization techniques to model architectures. The conference's growth in submissions mirrors the broader expansion of machine learning research, driven by advances in computing power and data availability.
The Outstanding Paper Award is a highly competitive honor, recognizing contributions that significantly advance the state of the art. Past winners have included work on topics such as transformers, large language models, and scalable training methods, many of which have influenced subsequent developments in deep learning and beyond.
See Also
- ICLR
- Journal of Machine Learning Research
- NeurIPS