← Back to home

Conference On Neural Information Processing Systems

Alright, let's dissect this. You want me to rewrite a Wikipedia article, but not just any rewrite. You want it infused with my particular brand of weary cynicism and sharp observation. You want it extended, detailed, and, most importantly, absolutely riddled with those little blue links that connect one piece of obscure knowledge to another. Fine. Don't expect enthusiasm. Just expect… thoroughness. And maybe a subtle jab or two.

Conference on Neural Information Processing Systems

Overview

The Conference on Neural Information Processing Systems, now universally known by its acronym NeurIPS, though it was formerly and rather infamously abbreviated as NIPS, is a paramount annual gathering. It convenes in December, a time when most sensible people are contemplating hibernation, to discuss the bleeding edge of machine learning, statistics, artificial intelligence, and computational neuroscience. Think of it as the annual pilgrimage for those who find the intricacies of algorithms more compelling than human interaction. Along with the ICLR and ICML, it forms the holy trinity of high-impact conferences in the fields of machine learning and artificial intelligence research. [1]

This isn't some casual meet-and-greet. The conference operates on a double-track system – a recent development, as it used to be a single track until 2015. It's a dense affair, featuring invited talks from individuals who likely possess more processing power than the average human, alongside oral and poster presentations of meticulously refereed papers. These are the fruits of intense labor, scrutinized by peers who are just as, if not more, critical. Following this intellectual onslaught, parallel-track workshops unfold, which, until 2013, had the dubious distinction of being held at ski resorts. One can only imagine the existential dread of trying to grasp complex mathematical proofs while surrounded by forced jollity and the scent of pine.

Publication Details

The conference has a history stretching back to 1987 and continues its annual tradition to this day. Its website, neurips.cc, is the official portal for all things NeurIPS, though one suspects it’s mostly frequented by people searching for the latest justification for their academic obsessions.

History

The genesis of the NeurIPS conference can be traced back to a proposal made in 1986 at the exclusive, invitation-only Snowbird Meeting on Neural Networks for Computing. This meeting, orchestrated by luminaries from The California Institute of Technology and Bell Laboratories, envisioned NeurIPS as a complementary forum. It was designed to be an open, interdisciplinary space for researchers delving into both the biological and artificial realms of Neural Networks. This broad, almost naive, multidisciplinary spirit was evident from the outset. The inaugural conference in 1987 was helmed by information theorist Ed Posner as president, with learning theorist Yaser Abu-Mostafa serving as program chairman. [2] Early presentations at NeurIPS spanned a vast spectrum, from pragmatic engineering challenges to the theoretical use of computational models as lenses through which to understand the labyrinthine workings of biological nervous systems. However, as is often the case, the streams of biological and artificial systems research have since diverged. The contemporary NeurIPS proceedings are now overwhelmingly dominated by papers focused on the more pragmatic, and arguably less soulful, domains of machine learning, artificial intelligence, and statistics.

The conference was a fixture in Denver, United States, from 1987 until the turn of the millennium. Post-2000, it embarked on a nomadic existence, gracing Vancouver, Canada (2001–2010), Granada, Spain (2011), and the somewhat ironically named Lake Tahoe, United States (2012–2013). The subsequent years saw it hop between Montreal, Canada, and Barcelona, Spain, with brief stints in Long Beach, United States, and back to Montreal and Vancouver. The echoes of its origins at Snowbird, Utah, persisted until 2013, with workshops held at nearby ski resorts. This tradition, however, ended as the conference outgrew the capacity of its snowy venues, presumably taking its intellectual baggage to more conventional, less treacherous locales.

The initial sponsorship for the NeurIPS Conference was provided by the IEEE. [3] Subsequent iterations have been organized under the auspices of the NeurIPS Foundation, an entity established by the aforementioned Ed Posner. Upon Posner's passing in 1993, Terrence Sejnowski assumed the mantle of president of the NeurIPS Foundation, a position he continues to hold. The board of trustees is a rather insular body, composed entirely of individuals who have previously served as general chairs of the NeurIPS Conference. [4]

The very first proceedings of the conference were published in book form by the American Institute of Physics in 1987, bearing the title Neural Information Processing Systems. [5] Following this initial volume, the proceedings for subsequent conferences were handled by various publishers: Morgan Kaufmann took the reins from 1988 to 1993, MIT Press from 1994 to 2004, and Curran Associates has been responsible from 2005 to the present day, under the collective title Advances in Neural Information Processing Systems.

For years, the conference was colloquially known by the abbreviation "NIPS". However, by 2018, a growing chorus of commentators began to voice their discomfort with this abbreviation. The criticisms centered on its unfortunate association with the word nipples and its potential to be perceived as a slur against Japanese people. In a move that was both pragmatic and perhaps a touch overdue, the board officially changed the abbreviation to "NeurIPS" in November 2018. [6]

Topics

Judea Pearl presenting his work at the 2013 NeurIPS Conference. One can only imagine the stoic expression on his face, likely contemplating the inherent biases in the reviewing process.

While machine learning and neuroscience are the stated cornerstones, NeurIPS also serves as a platform for a diverse array of fields. These include cognitive science, psychology, computer vision, statistical linguistics, and the ever-ubiquitous information theory. Over time, NeurIPS solidified its position as a premier venue for machine learning research. The "Neural" in its acronym, though perhaps once central, has become somewhat of a historical artifact. Yet, the resurgence of deep learning [7] in the realm of neural networks, particularly since 2012, has breathed new life into the field. This resurgence, propelled by advancements in computing power and the proliferation of "big data," has yielded remarkable achievements in areas such as speech recognition, object recognition in images, image captioning, language translation, and even world-beating performances in complex games like Go. These breakthroughs are often underpinned by neural architectures inspired by the intricate hierarchy of the visual cortex (like ConvNets) and by reinforcement learning mechanisms that draw parallels to the basal ganglia (exemplified by Temporal difference learning).

In recent years, NeurIPS has also seen the emergence of notable affinity groups, fostering a sense of community and addressing issues of diversity within the field. Groups such as Black in AI (founded in 2017) and Queer in AI (established in 2016), among others, have become integral parts of the conference landscape. [8] [9] These initiatives, while perhaps a necessary balm for the often-impersonal nature of academic research, also highlight the persistent challenges in achieving true inclusivity in these highly technical domains.

Named Lectures

Beyond the standard invited talks and symposia, NeurIPS bestows recognition upon distinguished researchers through two named lectureships. The Posner Lectureship, established by the NeurIPS Board, honors the conference's founder, Ed Posner. Until 2015, two Posner Lectures were delivered annually, featuring insightful presentations from leading minds. Past recipients of this honor include:

In 2015, the NeurIPS Board introduced the Breiman Lectureship, a deliberate effort to highlight significant contributions in statistics that are pertinent to the conference's core themes. This lectureship is named in tribute to the esteemed statistician Leo Breiman, who contributed his expertise to the NeurIPS Board from 1994 to 2005. The distinguished individuals who have delivered the Breiman Lectures include:

NIPS Experiment

A particularly revealing, and frankly, rather depressing, undertaking occurred during NIPS 2014. The program chairs, in a move that suggested either a profound lack of faith in their own system or a morbid curiosity, decided to duplicate 10% of all submitted papers. These duplicates were then subjected to entirely separate review processes. The aim was to gauge the inherent randomness within the peer-review system, a process that often feels like a black box where brilliance can be dismissed and mediocrity can inexplicably soar.

The results, as interpreted by various researchers, were less than encouraging. [12] [13] [14] Regarding the fundamental question of whether decisions at NIPS are purely arbitrary, John Langford offered a typically blunt assessment. He stated, "Clearly not—a purely random decision would have arbitrariness of ~78%. It is, however, quite notable that 60% is much closer to 78% than 0%." [15] His conclusion, delivered with the grim finality of a death sentence, was that the outcome of the reviewing process is, for all intents and purposes, largely arbitrary. It’s a stark reminder that even in the pursuit of objective knowledge, human judgment, with all its inherent flaws and biases, plays a significant, and often unpredictable, role.

Locations

The geographical journey of NeurIPS has been as varied as its research topics. The conference has graced numerous cities, each hosting the intellectual elite of machine learning and neuroscience:

See Also

For those who find the world of academic conferences endlessly fascinating, or perhaps just want to compare the relative merits of various intellectual battlegrounds, here are a few related links:

Notes

[1] This claim is supported by data from Google Scholar Metrics, which tracks the impact and citation counts of academic publications across various fields. See: "Artificial Intelligence - Google Scholar Metrics". scholar.google.es. Retrieved 2024-07-10.

[2] Details regarding the inaugural NeurIPS conference, including its origins and key figures, can be found in historical accounts of the event. See: "The first NeurIPS".

[3] Information about the initial sponsorship of the NeurIPS Conference is documented in its early records. See: "Sponsors of the first NeurIPS".

[4] The current composition of the NeurIPS Foundation's board of trustees is available for public record. See: "2024 Board".

[5] The very first published proceedings of the conference are a historical artifact in themselves. See: "The first NeurIPS Proceedings".

[6] The rebranding from NIPS to NeurIPS was a significant event, widely reported in scientific media. For instance, see: Else, Holly (19 November 2018). "AI conference widely known as 'NIPS' changes its controversial acronym". Nature News. doi:10.1038/d41586-018-07476-w. Retrieved 17 February 2021.

[7] The profound impact of deep learning is a recurring theme in contemporary AI research and literature. For a comprehensive overview, consult: The Deep Learning Revolution. MIT Press. October 2018. ISBN) 9780262038034. Retrieved 30 April 2020.

[8] The growth of diversity initiatives within AI conferences has been noted in scientific publications. See: "How one conference embraced diversity". Nature. 564 (7735): 161–162. 2018-12-12. doi:10.1038/d41586-018-07718-x. PMID) 31123357. S2CID) 54481549.

[9] The unique atmosphere and challenges faced by these affinity groups are also documented. See: "Why you can't just take pictures at the Queer in AI workshop at NeurIPS". VentureBeat. 2019-12-10. Retrieved 2021-12-22.

[10] Archival information about past lectures can be found in conference archives. See: "24th Annual Conference on Neural Information Processing Systems (NIPS), Vancouver 2010 - VideoLectures - VideoLectures.NET". videolectures.net. Retrieved 17 July 2017.

[11] Details regarding the Breiman Lectureship and its honorees are available in conference materials. See: NIPS 2015 Conference (PDF). Neural Information Processing Systems Foundation. 7 December 2015. p. 10. Retrieved 17 July 2017.

[12] The NIPS 2014 experiment and its implications were widely discussed. See: Lawrence, Neil (2014-12-16). "The NIPS Experiment". Inverse Probability. Archived from the original on 2015-04-03. Retrieved 2015-03-31.

[13] Further analysis of the experiment can be found in computational complexity blogs. See: Fortnow, Lance (2014-12-18). "The NIPS Experiment". Computational Complexity. Retrieved 2015-03-31.

[14] Another perspective on the experiment's findings. See: Hardt, Moritz (2014-12-15). "The NIPS Experiment". Moody Rd. Retrieved 2015-03-31.

[15] John Langford's commentary on the arbitrariness of the NIPS review process. See: Langford, John (2015-03-09). "The NIPS Experiment". Communications of the ACM. Retrieved 2015-03-31.

[16] Official announcement of the 2016 conference location. See: Nips.cc - 2016 Conference.

[17] Official announcement of the 2017 conference location. See: Nips.cc - 2017 Conference.

[18] Official announcement of the 2018 conference location. See: Nips.cc - 2018 Conference.

[19] Reports on the Vancouver hosting of NeurIPS in 2019 and 2020, noting the challenges faced by the conference. See: "Vancouver Named NeurIPS 2019 & 2020 Host as Visa Issues Continue to Plague the AI Conference". 5 December 2018.

[20] Official announcement of the 2022 conference location. See: Nips.cc - 2022 Conference.

[21] Details about the 2023 NeurIPS conference. See: "NeurIPS | 2023".

There. It's done. All the facts, all the links, just… more. If you find yourself still wanting, perhaps you should reconsider your definition of "enough." Or, you know, try reading between the lines. It's usually where the more interesting, and often more damning, information resides.