The Conference on Neural Information Processing Systems (abbreviated as NeurIPS, formerly NIPS) is a Machine learning and computational neuroscience conference held annually in December. Along with ICLR and ICML, it is one of the three primary high-impact conferences in Machine learning and Artificial intelligence research. The conference includes three days of invited talks along with oral and poster presentations of refereed papers, followed by two days of workshops and competitions.
The NeurIPS meeting was first proposed in 1986 at the annual invitation-only Snowbird Meeting on Neural Networks for Computing, organized by the California Institute of Technology and Bell Laboratories. It was conceived as an open interdisciplinary gathering for researchers studying biological and artificial neural networks, complementing the private Snowbird event. The first conference took place in 1987 in Denver, Colorado, with information theorist Ed Posner as conference president and learning theorist Yaser Abu-Mostafa as program chairman.
Founding and early years
The founding of NeurIPS in 1987 marked a deliberate departure from the invitation-only format of its predecessor. The inaugural meeting was sponsored by the IEEE and attracted researchers from diverse fields including engineering, neuroscience, and statistics. Early proceedings reflected a broad spectrum of work, ranging from purely engineering-oriented problems to computational models of biological nervous systems. As the fields of Machine learning and Artificial intelligence matured, the biological and engineering research streams diverged, and by the 2000s the conference program had become dominated by machine learning, statistics, and AI research.
Ed Posner, an information theorist at the California Institute of Technology, served as the first conference president, and Yaser Abu-Mostafa acted as the program chairman. The conference was initially held in Denver, Colorado, from 1987 through 2000. Afterward it moved to Vancouver (2001–2010), then Granada (2011), and Lake Tahoe (2012–2013), before rotating among several international cities in subsequent years.
Organizational History
The First NeurIPS conference was sponsored by the IEEE. Subsequent meetings were organized by the NeurIPS Foundation, which was established by Ed Posner. Terrence Sejnowski has served as president of the foundation since Posner's death in 1993. The board of trustees is composed of previous general chairs of the conference.
Conference proceedings were first published in book form by the American Institute of Physics in 1987 as Neural Information Processing Systems. Subsequent volumes were released by Morgan Kaufmann (1988–1993), MIT Press (1994–2004), and Curran Associates (2005–present) under the title Advances in Neural Information Processing Systems.
Naming and Renaming
The conference was originally abbreviated as "NIPS". In 2018, commentators criticized the acronym for its association with the word "nipples" and argued it could encourage sexism within the field. In response, the organizing board changed the official abbreviation to "NeurIPS" in November 2018.
Conference Format and Scope
NeurIPS is held annually and includes invited talks, oral presentations, and poster sessions for refereed papers, followed by workshops and competitions. Alongside ICLR and ICML, it is considered a top-tier venue for publishing research in machine learning and artificial intelligence.
The topical scope of the conference has expanded from its original focus on neural networks to include cognitive-science, Computer vision, Natural language processing, and statistics. The resurgence of Deep learning since around 2012 has positioned NeurIPS as a central venue for advances in neural network architectures and reinforcement learning, including breakthroughs in image recognition, language translation, and game-playing systems.
Named Lectures
NeurIPS organizes two named lectureships honoring distinguished researchers. The Posner Lectureship, introduced by the NeurIPS Board, honors founder Ed Posner. Past Posner lecturers include:
- 2010: Michael Jordan and Rich Sutton
- 2011: Thomas Dietterich and Terry Sejnowski
- 2012: Daphne Koller and Peter Dayan
- 2013: Yoshua Bengio and Zoubin Ghahramani
- 2014: Yann LeCun and Peter Dayan
- 2015: Zoubin Ghahramani and Yoshua Bengio
- 2016: Michael Kearns
- 2017: John Platt
- 2018: Joelle Pineau
- 2019: Yoshua Bengio
- 2020: Peter Bartlett
The Breiman Lectureship, introduced in 2015, honors statistician Leo Breiman, who served on the NeurIPS board from 1994 to 2005. Past lecturers include Susan Holmes (2016), David Spiegelhalter (2018), Emmanuel Candes (2022), and Arnaud Doucet (2024).
Review Consistency Experiments
In 2014, program chairs duplicated 10% of all submissions and sent them to separate reviewers, a test of the peer-review process. Results showed that around 60% of decisions were consistent, which several researchers interpreted as evidence of substantial arbitrariness in the review outcome. John Langford noted that a purely random process would yield roughly 78% inconsistency, so the observed 60% was closer to random than to perfect agreement. The experiment was repeated at NeurIPS 2021, with similar findings: 23% of duplicated submissions received different accept/reject decisions, and 50.6% of accepted papers would have been rejected under a different review group.
Impact and Legacy
Over its decades-long history, NeurIPS has become one of the three primary conferences of high impact in Artificial intelligence research. The resurgence of Deep learning since the early 2010s, enabled by faster hardware and larger datasets, has made the conference a central venue for advances in computer vision, natural language processing, and reinforcement learning. NeurIPS has also played a role in fostering community diversity through affinity groups such as Black in AI (2017) and Queer in AI (2016).
Named Lectureships
The NeurIPS Board established the Posner Lectureship in honor of founder Ed Posner; it was presented to two honorees per year up to 2015. Recipients have included Josh Tenenbaum (2012), Bernhard Schölkopf (2017), and Yann LeCun (2016). The Breiman Lectureship was introduced in 2015, named after statistician Leo Breiman, and has been awarded to researchers including Emmanuel Candes (2022) and Arnaud Doucet (2024).
Peer Review Consistency
In 2014, program chairs conducted a consistency experiment by duplicating 10% of submissions and sending them through separate reviewers. The results indicated significant arbitrariness in the review process, with an estimated inconsistency of around 60%, compared to the roughly 78% expected from a purely random decision process. A follow-up experiment at NeurIPS 2021 found that 23% of duplicated submissions received different accept/reject decisions, and over half of accepted papers would have been rejected under re-review.
Locations and Growth
From 1987 to 2000, the conference met in Denver. It then moved to Vancouver (2001–2010), Granada (2011), Lake Tahoe (2012–2013), Montréal (2014–2015), Barcelona (2016), Long Beach (2017), Montréal (2018), and Vancouver (2019). The 2020 and 2021 editions were held virtually due to the COVID-19 pandemic. Subsequent venues included New Orleans (2022), New Orleans (2023), and Vancouver (2024). Future planned locations include San Diego (2025) and Sydney (2026).