International Conference on Machine Learning (ICML)

What ICML is

The International Conference on Machine Learning (ICML) is an annual academic conference devoted to machine learning. It has been held every year since 1980, which makes it the oldest conference in the field and, together with NeurIPS and ICLR, one of the three venues with the highest impact and reputation in machine learning and artificial intelligence research^[1]. It is organized by the International Machine Learning Society (IMLS)^[1]. Its official website describes it as the premier gathering of professionals dedicated to advancing the branch of artificial intelligence known as machine learning, and as one of the fastest-growing artificial intelligence conferences in the world^[2].

How it started

ICML began as the International Workshop on Machine Learning. The first meeting was held at Carnegie Mellon University in Pittsburgh in July 1980, organized by Jaime Carbonell, Ryszard S. Michalski and Tom M. Mitchell^[1]. It ran as a workshop series before transitioning, in 1993, into a full conference series under its current name, building on the earlier workshops^[1].

What it covers

Its scope spans the full breadth of machine learning. Compared with the other two major machine-learning conferences, ICML has traditionally carried more content on statistical learning theory, reinforcement learning and robotics, and optimization theory^[1]. The official website frames the programme around research on all aspects of machine learning and its use in closely related areas such as artificial intelligence, statistics and data science, as well as in application areas including machine vision, computational biology, speech recognition and robotics^[2].

Growth in numbers

The conference's growth is visible in its submission figures. Submissions rose from 1,037, with 270 accepted, in 2015 to 12,107, with 3,260 accepted, in 2025, while average acceptance rates over that period ranged between 21% in 2020 and 30% in 2024^[1]. In 2026 the conference received a record 24,371 submitted papers, of which 2.2% were designated as spotlight and 0.7% as oral presentations^[1].

Where the papers go

Since the 2010s, ICML's proceedings have been published open access in the Proceedings of Machine Learning Research (PMLR), which has carried that name since 2015^[1]. The official proceedings are hosted at proceedings.mlr.press, which is linked from the conference's website^[2]. Work presented at the conference includes Google's Batch Normalization, Google Brain's EfficientNet (ICML 2019), OpenAI's CLIP (ICML 2021) and UC Berkeley's Soft Actor-Critic (ICML 2018)^[1].

Where it sits in the field

ICML's position comes from a combination of age, scale and breadth. Having started as a workshop in 1980 and become a full conference in 1993, it has been running longer than any other meeting in machine learning, and it now reviews a volume of submissions in the tens of thousands^[1]. Its programme reaches from statistical learning theory and optimization to robotics and application areas such as computational biology and speech recognition, and its papers are freely available through PMLR^[1][2]. For readers trying to follow where machine learning research is going in a given year, the conference's proceedings are one of the standard places to look.