The International Conference on Learning Representations (ICLR) is a prominent academic conference in the field of machine learning, first held in 2013. It is dedicated to research on learning representations, particularly deep learning and neural networks. Along with NeurIPS and ICML, ICLR is considered one of the three primary conferences of highest impact and reputation in machine learning and artificial intelligence research.
The conference was founded by Yann LeCun and Yoshua Bengio in 2012, with the inaugural event taking place in 2013. From its inception, ICLR adopted an open peer review process for paper submissions, a model proposed by LeCun. This approach makes all submitted papers and reviews publicly available online, contrasting with the traditional double-blind review used by many other conferences.
Conference Format
ICLR is typically held in late April or early May each year. The program includes invited talks, oral presentations, and poster sessions of refereed papers. The open review process is a defining feature, allowing the community to comment on submissions before acceptance decisions are made. This transparency is intended to foster faster dissemination of ideas and more collaborative scientific discourse.
The conference covers a broad range of topics, including but not limited to: unsupervised and semi-supervised learning, generative models, optimization methods such as the Adam optimizer and other SGD variants, regularization techniques like dropout and batch normalization, and architectures such as transformers and residual networks.
Founding and Motivation
The founding of ICLR was motivated by the rapid growth of interest in deep learning in the early 2010s. LeCun and Bengio, both pioneers in the field, sought to create a venue that would focus specifically on learning representations, a topic they felt was underserved by existing conferences like NeurIPS and ICML. The open review model was a key innovation, designed to speed up the review cycle and make the process more democratic.
The first ICLR in 2013 attracted a relatively small but highly engaged community. Over the years, the conference has grown substantially, becoming one of the most selective and influential venues in the field. Its acceptance rate has consistently been low, reflecting the high quality of submissions.
Impact and Reputation
ICLR is now widely regarded as a top-tier conference, with papers published there often having significant impact on both academia and industry. Many foundational works in deep learning, including developments in large language models and multi-head attention, have been presented at ICLR. The conference's open review process has also influenced other venues, with some adopting similar models.
The conference is supported by major technology companies and research institutions, including Google DeepMind, OpenAI, Anthropic, and the University of Toronto, among others. This industry involvement underscores the close connection between ICLR research and practical applications in areas like natural language processing, computer vision, and generative AI.
Editions and Growth
Since 2013, ICLR has been held annually. The location has varied, with editions in the United States, Canada, Europe, and Asia. The conference has consistently grown in terms of submissions and attendance, reflecting the expanding scope of machine learning research. As of the early 2020s, ICLR receives thousands of submissions each year, making it one of the largest and most competitive conferences in the field.
The conference also includes workshops and tutorials, providing additional opportunities for researchers to present work-in-progress and discuss emerging topics. The open review process extends to these ancillary events, maintaining the conference's commitment to transparency.