2025 Conference

The NeurIPS Journal Track aims to strengthen the bridge between machine learning and statistics communities and enable the dissemination of scientific knowledge. Journals such as JMLR and AoS accept rigorously reviewed papers on a rolling schedule outside of the conference cycle. This track allows the top accepted papers to also be presented at NeurIPS to get increased visibility and spark discussion.

This year, we are thrilled to feature 34 accepted papers — 14 from the Journal of Machine Learning Research (JMLR) and 20 from the Annals of Statistics (AoS) — all of which will be presented as posters at NeurIPS 2025 in San Diego. These papers represent the highest standards of scholarship across theory, methodology, and applications, and they exemplify how journal-based research continues to enrich the broader NeurIPS community. 

We invite all attendees to visit the Journal Track posters, meet the authors, and engage in discussions that highlight the evolving connections between machine learning and statistics.

---

Accepted Papers from the Journal of Machine Learning Research (JMLR)

1) RLtools: A Fast, Portable Deep Reinforcement Learning Library for Continuous Control

Jonas Eschmann, Dario Albani, Giuseppe Loianno

https://www.jmlr.org/papers/v25/24-0248.html

2) Bridging Distributional and Risk-sensitive Reinforcement Learning with Provable Regret Bounds

Hao Liang, Zhi-Quan Luo

https://www.jmlr.org/papers/v25/22-1253.html

3) Learning with Norm Constrained, Over-parameterized, Two-layer Neural Networks

Fanghui Liu, Leello Dadi, Volkan Cevher

https://jmlr.org/papers/v25/22-1250.html

4) Uniform Generalization Bounds on Data-Dependent Hypothesis Sets via PAC-Bayesian Theory on Random Sets

Benjamin Dupuis, Paul Viallard, George Deligiannidis, Umut Simsekli

https://www.jmlr.org/papers/v25/24-0605.html

5) Stationary Kernels and Gaussian Processes on Lie Groups and their Homogeneous Spaces I: the compact case

Iskander Azangulov, Andrei Smolensky, Alexander Terenin, Viacheslav Borovitskiy

https://jmlr.org/papers/v25/22-1434.html

6) Stationary Kernels and Gaussian Processes on Lie Groups and their Homogeneous Spaces II: non-compact symmetric spaces

Iskander Azangulov, Andrei Smolensky, Alexander Terenin, Viacheslav Borovitskiy

https://jmlr.org/papers/v25/23-0115.html

7) Dropout Regularization Versus l2-Penalization in the Linear Model

Gabriel Clara, Sophie Langer, Johannes Schmidt-Hieber

https://www.jmlr.org/papers/v25/23-0803.html

8) Interpretable Global Minima of Deep ReLU Neural Networks on Sequentially Separable Data

Thomas Chen, Patrícia Muñoz Ewald

https://jmlr.org/papers/v26/24-1516.html

9) Stochastic-Constrained Stochastic Optimization with Markovian Data

Yeongjong Kim, Dabeen Lee

https://jmlr.org/papers/v25/23-1630.html

10) Geometric Learning with Positively Decomposable Kernels

Nathaël Da Costa, Cyrus Mostajeran, Juan-Pablo Ortega, and Salem Said

https://jmlr.org/papers/v25/23-1400.html

11) The ODE Method for Stochastic Approximation and Reinforcement Learning with Markovian Noise

Shuze Daniel Liu, Shuhang Chen, Shangtong Zhang

https://www.jmlr.org/papers/volume26/24-0100/24-0100.pdf

12) On the Convergence of Projected Policy Gradient for Any Constant Step Sizes

Jiacai Liu, Wenye Li, Dachao Lin, Ke Wei, Zhihua Zhang

https://www.jmlr.org/papers/volume26/24-1530/24-1530.pdf

13) Data-Driven Performance Guarantees for Classical and Learned Optimizers

Rajiv Sambharya, Bartolomeo Stellato

https://jmlr.org/papers/volume26/24-0755/24-0755.pdf

14) Fine-grained Analysis and Faster Algorithms for Iteratively Solving Linear Systems

Michal Dereziński, Daniel LeJeune, Deanna Needell, Elizaveta Rebrova

https://jmlr.org/papers/volume26/24-1906/24-1906.pdf

Accepted Papers from the Annals of Statistics (AoS)

1) Versatile Differentially Private Learning for General Loss Functions

Song X Chen, Qilong Lu, Yumou Qiu

https://www.e-publications.org/ims/submission/AOS/user/submissionFile/71306?confirm=2c6478db

2) Distributionally Robust Learning for Multi-source Unsupervised Domain Adaptation

Zhenyu Wang, Peter Bühlmann, Zijian Guo

https://www.e-publications.org/ims/submission/AOS/user/submissionFile/69288?confirm=dbfbb000

3) Neural Networks Generalize on Low Complexity Data

Sourav Chatterjee and Timothy Sudijono

https://www.e-publications.org/ims/submission/AOS/user/submissionFile/67274?confirm=7d909619

4) Pseudo-Labeling for Kernel Ridge Regression under Covariate Shift

Kaizheng Wang

https://www.e-publications.org/ims/submission/AOS/user/submissionFile/59591?confirm=539ffe2f

5) A Geometrical Analysis of Kernel Ridge Regression and its Applications

Zong Shang, Guillaume Lecué, Georgios Gavrilopoulos

https://www.e-publications.org/ims/submission/AOS/user/submissionFile/66957?confirm=79d1f75f

6) Improved Learning Theory for Kernel Distribution Regression with Two-Stage Sampling

François Bachoc, Louis Béthune, Alberto González-Sanz, Jean-Michel Loubes

https://projecteuclid.org/journals/annals-of-statistics/volume-53/issue-4/Improved-learning-theory-for-kernel-distribution-regression-with-two-stage/10.1214/25-AOS2537.short

7) Robust Transfer Learning with Unreliable Source Data

Jianqing Fan, Cheng Gao, Jason M. Klusowski

https://projecteuclid.org/journals/annals-of-statistics/volume-53/issue-4/Robust-transfer-learning-with-unreliable-source-data/10.1214/25-AOS2534.short

8) Online Estimation and Inference for Robust Policy Evaluation in Reinforcement Learning

Weidong Liu, Jiyuan Tu, Yichen Zhang, Xi Chen

https://www.e-publications.org/ims/submission/AOS/user/submissionFile/66491?confirm=9350ae31

9) Estimation and Inference in Distributional Reinforcement Learning

Liangyu Zhang, Yang Peng, Jiadong Liang, Wenhao Yang, Zhihua Zhang

https://www.e-publications.org/ims/submission/AOS/user/submissionFile/66608?confirm=25fa0a06

10) Online Statistical Inference in Decision Making with Matrix Context

Qiyu Han, Will Wei Sun, Yichen Zhan

https://www.e-publications.org/ims/submission/AOS/user/submissionFile/64262?confirm=82a6f9e7

11) Asymptotic Theory of Geometric and Adaptive $k$-Means Clustering

Adam Quinn Jaffe

https://projecteuclid.org/journals/annals-of-statistics/volume-53/issue-4/Asymptotic-theory-of-geometric-and-adaptive-k-means-clustering/10.1214/25-AOS2514.short

12) Reinforcement Learning for Individual Optimal Policy from Heterogeneous Data

Rui Miao, Babak Shahbaba, Annie Qu

https://projecteuclid.org/journals/annals-of-statistics/volume-53/issue-4/Reinforcement-learning-for-individual-optimal-policy-from-heterogeneous-data/10.1214/25-AOS2512.full

13) Policy Larning “without” overlap: Pessimism and Generalized Empirical Bernstein’s Inequality

Ying Jin, Zhimei Ren, Zhuoran Yang, Zhaoran Wang

https://projecteuclid.org/journals/annals-of-statistics/volume-53/issue-4/Policy-learning-without-overlap–Pessimism-and-generalized-empirical-Bernsteins/10.1214/25-AOS2511.short

14) Testing Stationarity and Change Point Detection in Reinforcement Learning

Mengbing Li, Chengchun Shi, Zhenke Wu, Piotr Fryzlewicz

https://projecteuclid.org/journals/annals-of-statistics/volume-53/issue-3/Testing-stationarity-and-change-point-detection-in-reinforcement-learning/10.1214/25-AOS2501.short

15) A Duality Framework for Analyzing Random Feature and Two-Layer Neural Networks

Hongrui Chen, Jihao Long, Lei Wu

https://projecteuclid.org/journals/annals-of-statistics/volume-53/issue-3/A-duality-framework-for-analyzing-random-feature-and-two-layer/10.1214/25-AOS2492.short

16) Unified Algorithms for RL with Decision-Estimation Coefficients: PAC, Reward-Free, Preference-Based Learning, and Beyond

Fan Chen, Song Mei, Yu Bai

https://projecteuclid.org/journals/annals-of-statistics/volume-53/issue-1/Unified-algorithms-for-RL-with-Decision-Estimation-Coefficients–PAC/10.1214/24-AOS2483.full

17) Multivariate Dynamic Mediation Analysis under a Reinforcement Learning Framework

Lan Luo, Chengchun Shi, Jitao Wang, Zhenke Wu, Lexin Li

https://projecteuclid.org/journals/annals-of-statistics/volume-53/issue-1/Multivariate-dynamic-mediation-analysis-under-a-reinforcement-learning-framework/10.1214/24-AOS2475.short

18) A Statistical Framework of Watermarks for Large Language Models: Pivot, Detection Efficiency and Optimal Rules

Xiang Li, Feng Ruan, Huiyuan Wang, Qi Long, Weijie J. Su

https://projecteuclid.org/journals/annals-of-statistics/volume-53/issue-1/A-statistical-framework-of-watermarks-for-large-language-models/10.1214/24-AOS2468.short

19) Statistical Inference for Decentralized Federated Learning

Jia Gu, Song Xi Chen

https://projecteuclid.org/journals/annals-of-statistics/volume-52/issue-6/Statistical-inference-for-decentralized-federated-learning/10.1214/24-AOS2452.short

20) Deep Nonlinear Sufficient Dimension Reduction

YinFeng Chen, YuLing Jiao, Rui Qiu, Zhou Yu

https://projecteuclid.org/journals/annals-of-statistics/volume-52/issue-3/Deep-nonlinear-sufficient-dimension-reduction/10.1214/24-AOS2390.short

---

These presentations highlight the growing collaboration between NeurIPS and the leading journals that define our field. We look forward to welcoming all authors and attendees to San Diego this December for an inspiring exchange of ideas that bridges deep theory, rigorous methods, and transformative applications.

NeurIPS 2025 Journal Chair

Lam Nguyen

We are excited to announce this year’s NeurIPS Affinity Events! 

What are affinity groups?

At NeurIPS, affinity groups play a crucial role in promoting and supporting the ideas and voices of various communities that are defined by some axis of joint identity and raise awareness of issues that affect their members. In addition, they provide members of these affinity groups with increased opportunities to showcase their work, engage in discussions, and build connections during NeurIPS events, promoting diversity and inclusion at NeurIPS.

The main goals of affinity events are to 1) increase visibility of the work done by the members of the affinity group, 2) enable them to interact and discuss topics that are unique to each affinity group, and 3) encourage affinity group members to showcase their work and step into the NeurIPS community.  Affinity groups organize events with the main aim to encourage people to showcase their work and step into the NeurIPS community. 

What kind of events do affinity groups offer?

In contrast to regular workshops, affinity group workshops adopt a more open and flexible approach around topic areas, focusing on forming research communities within the affinity groups and bringing attention to the unique challenges faced by their members.

Additionally, affinity group socials serve as safe and welcoming spaces, allowing members to interact, socialize, and network in an inclusive environment. By fostering such social gatherings, NeurIPS supports the creation of a supportive community that values diversity and ensures that all voices are heard and valued. 

The affinity events at NeurIPS this year will take place throughout the conference week in San Diego. Moreover, there will be a Joint Affinity Event organized by Queer in AI, Latinx in ML and Women in ML in Mexico City.

What can NeurIPS attendees do to support affinity groups?

Participate! Actively attending and participating in panels and sessions for discussions about issues that are relevant to the affinity groups goes a long way in helping promote their cause. We request all members of our community to participate actively!

Affinity groups at NeurIPS 2025

We are delighted to announce the following affinity groups at NeurIPS 2025. Find out more below about their mission and the event they are organizing! 

Indigenous in AI/ML 

Indigenous in AI/ML is a networking and support group to connect Indigenous professionals and researchers working in artificial intelligence and machine learning. Our goal is to amplify the voices of Native, Aboriginal, and First Nations experts, inspire impactful work that benefits Indigenous communities, and challenge stereotypes in advanced technology fields.

In addition to fostering connections locally, regionally, and globally, our group educates the broader tech community about contemporary Indigenous issues relevant to AI and information technology. By cultivating an international network of Indigenous AI/ML professionals who inspire innovation and foster community empowerment, we can shape the future of technology through Indigenous knowledge and perspectives. The workshop will take place on Wednesday, December 3. For more details, please visit: www.indigigenius.org. 

LatinX in AI 

The LXAI Workshop has been a recognized part of the NeurIPS workshop lineup since 2018. This one-day event is designed to bring together faculty, students, researchers, and professionals for a day of collaboration and knowledge-sharing. The agenda includes invited talks from leading experts, oral and poster presentations of cutting-edge research, a panel discussion on key topics in the field, and a dedicated mentoring session.

Each year, we receive high-quality paper submissions from Latinx researchers worldwide, demonstrating the community’s growing strength and interest in AI. We aim to amplify Latinx voices and promote diverse perspectives to build a more inclusive and equitable AI research ecosystem. While our focus is on Latinx representation, the event is intentionally inclusive and open to all who are committed to advancing diversity in AI.

While all presenters will identify primarily as LatinX, everyone is invited to attend. The workshop will take place on Tuesday, December 2. For more details, please visit: https://www.latinxinai.org/neurips-2025 

Muslims in ML 

The 5th Muslims in ML (MusIML) workshop will be held at NeurIPS 2025, bringing together a vibrant community at the intersection of machine learning, fairness, and global Muslim expertise. This year’s program will feature research by Muslim scholars on Muslim communities and contexts, as well as competition proposals for social impact. With the panel theme on Responsible AI, we will explore the ethical, societal, and cultural dimensions shaping the future of intelligent systems. Attendees can also look forward to hearing from top speakers and thought leaders from across the Muslim AI/ML community, sharing cutting-edge ideas and experiences. The workshop continues MusIML’s mission to amplify diverse voices, provide mentorship and collaboration opportunities, and build AI that truly reflects and serves global communities. The workshop will take place on Tuesday, December 2. For more details, visit our website: musiml.org/events/2025-NeurIPS

New in ML

In an era deeply influenced by the rapid rise of Large Language Models (LLMs) and Generative AI, how can researchers effectively navigate their journey in machine learning and accelerate their growth to achieve impactful breakthroughs? The NewInML Workshop brings together emerging researchers and senior PhD students, connecting them directly with top experts and industry thought-leaders for profound insights and practical guidance. Covering topics from foundational skills to advanced research in LLMs and Generative AI, the workshop offers a unique opportunity to enhance your research capabilities and competitive edge. Join us to ride the wave of innovation and grow alongside the brightest minds in the ML community! NewInML will be held Dec 2nd. For more details, visit our official website: newinml.github.io

Queer in AI 

The Queer in AI workshop and social events at NeurIPS 2025 aim to create a vibrant and inclusive space that brings together queer students, researchers, and practitioners from across the globe to critically examine the impact of AI on the LGBTQIA* community. The workshop aspires to build a strong network of queer individuals, enable discussions at the intersection of AI and queer identities, and tackle issues of discrimination within the tech industry and society at large.

The Queer in AI workshop is dedicated to advancing equitable AI practices and contributing to a future where technology serves the needs of all communities. The workshop is committed to advocating for a more ethically grounded approach to AI development, ensuring that diverse voices are heard and valued in conversations about the future of technology. 

For more details, please visit the our official website: https://www.queerinai.com/neurips-2025 

Women in ML 

The WiML workshop celebrates the research of women and non-binary scholars in machine learning. Our goals are to:

(1) Highlight their contributions across the breadth of ML

(2) Create spaces to meet, exchange ideas through deep technical discussion, and spark future collaborations

(3) Provide junior researchers with a visible platform to present their work to peers and senior members of the community.

The program will feature invited talks by women and non-binary researchers, complemented by structured mentorship sessions focused on practical topics such as navigating publishing, shaping research agendas, and exploring academic and industry paths. Dedicated networking time will help attendees build lasting connections and identify collaboration opportunities. Accepted abstracts will be presented in a poster session, showcasing a wide range of projects and fostering constructive, interactive feedback.

Our workshop is designed to be welcoming to researchers at all stages and across subfields.

The workshop will take place on Tuesday, December 2. 

For more details, please visit the WiML website and our 2025 workshop website.

NeurIPS 2025 Affinity Chairs

Ioana Bica, Nezihe Merve Gürel, and Sarthak Arora

As we announce decisions for the first-ever NeurIPS Position Paper Track, we find ourselves reflecting on an extraordinary experimental year. When NeurIPS launched this pilot track, we ventured into uncharted territory. The community response was overwhelming and inspiring. Every submission represented not just research, but a vision for how position papers could contribute to broader conversations in our field.

We want to take a moment to share our reflections, including lessons we’ll take forward into the future. This blog post is specific to the position paper track, which is just one of several tracks at NeurIPS: the Main Program track and Datasets and Benchmarks track are also releasing blog posts reflecting on their processes. To all authors who trusted us with their ideas and hard work, thank you. To all Reviewers, Emergency Reviewers, and Area Chairs who dedicated their time, thank you, too! 

The Numbers

We received nearly 700 initial submissions—a response that both humbled and energized us. After accounting for withdrawals and desk reviews, we had 496 viable candidates for full consideration. From this impressive pool, we selected 40 papers for presentation at the conference. That’s an acceptance rate of approximately 8%. We know that this number is lower than other tracks in the conference, but it reflects the specific affordances of position papers and the conference strategy for piloting this track. 

The NeurIPS 2025 Position Paper Track envisions giving each accepted author more focused attention: we are planning to provide accepted authors with new formats of presentation and tailored communication guidance to further evolve their message. We feel this better fits with the NeurIPS position paper track goal for this year of widespread visibility to serve as a springboard for public discussion. This does mean that we had to accept a limited number of papers.

If your paper received high scores but it wasn’t ultimately accepted for presentation at the conference, you’re in good company. We had a wealth of great papers this year! Additionally, some excellent papers with high technical merit weren’t the best fit for this inaugural community discussion—either requiring specialized expertise that might limit broader engagement or being more focused on technical implementations and methods than we envision for this year’s track.

We encourage authors who received positive reviews to incorporate that feedback for submissions to other venues and platforms. These conversations are too important to end here.

Clarifying Our Process

We’ve received questions about our process, and we want to be transparent as we continue to adapt and learn from this pilot year. We’re grateful for all the suggestions authors provided in the author survey, which included proposals for new review mechanisms, specific feedback on the call for submissions, and thoughtful ideas about track themes. We will continue to digest your proposals and translate these ideas into learnings for future conference position paper tracks.

One thing we piloted this year that we want to provide extra context on is the adjudication process. From the outset, adjudication was designed to improve our review process and address procedural concerns about reviewer behavior—not to alter paper scores after the fact. That said, all author feedback, including adjudication requests, was available to Area Chairs and Program Chairs to inform metareviews and decision-making.

We have several ongoing investigations into reviewer conduct concerns that will continue until appropriate resolution. For every author who requested adjudication, we’re corresponding individually.

Addressing Community Concerns

We’ve received thoughtful feedback that has highlighted some unintended consequences of our pilot approach. We want to address these directly:

Exclusivity and Limited Discussion: Our small acceptance pool, while intended to provide focused attention to selected papers, has understandably raised concerns about exclusivity and limited diversity of viewpoints. We stand by our decision to focus on fewer papers this year towards a goal of more amplification and resources on each, and the community feedback is something that we will take into account for future years.

Responses to Reviews: We experimented with changes to the review process, moving from rebuttals to an author survey and creating a space for adjudication where possible misbehavior might be in question. We were hoping to create more civil exchanges and provide mechanisms for addressing review concerns within the more constrained timeline of the position paper track. Our decision was Informed by a growing body of research that shows that the process of rebuttals can be biased and expose reviewers to “peer review bullying” Author survey responses, including how the paper would change based on reviewer feedback, were read and considered carefully in metareviews and the final decisions. In other words, we looked at everything when making the decisions.

Timeline Challenges: We apologize that our ambitious design was not accompanied by punctual logistics execution when we encountered storms along the way. We initially synchronized with other track deadlines within NeurIPS. Unfortunately, we found ourselves needing a significant number of emergency reviewers and additional time to ensure every paper received proper evaluation and a minimum of three reviews. This was a result of the novel format of the position paper track, uncertainty around reviewer expectations, and a greater number of reviewers who were unable to complete their work or became unresponsive than we had planned for. 

We did not anticipate the scale of reviewer coordination required and failed to meet our original timeline commitments. When it became clear this would happen, we sent an email to authors suggesting that they prepare their papers for re-submission to other conferences before our decisions came out. In future iterations of our track, we will work to plan deadlines more strategically. 

Lessons Learned

This pilot year has been a learning experience, including some challenging community feedback that prompted important reflection.

What We’ve Learned:

• We need to provide much clearer guidance about appropriate submissions and review expectations including examples that people can reference;
• We need to overrecruit and overassign reviewers to papers at a higher rather than dis-similar tracks to ensure a minimum of three reviews for every paper by the deadline; 
• The position paper format works best when facilitating broad, meaningful discussions.

Addressing Community Concerns:

• On Exclusivity: Our small acceptance pool raised concerns about limited diversity of viewpoints. While we have a small number of papers, we believe they touch on many of the topics we saw voiced by the community. We believe future position paper tracks should be resourced to target a higher acceptance rate so that more authors have the opportunity to present.
• On Review Process: Our track’s different processes were difficult for some authors who followed prior venue guidance for improving paper acceptance, meanwhile some authors and reviewers took the changes at face value. Future chairs should be more creative with the strategies we use not only to communicate our ideas, but to listen to feedback and constructive criticism.
• On Timeline Challenges: We needed more time to recruit, assign, and complete emergency reviews. This resulted in a final decision date that overlapped with other conferences. In future years, chairs should be more cognizant of deadlines for our sister conferences and try to provide further advance notice to authors on how to handle the conflict. 

We put out blog posts and communicated via OpenReview for our decision-making standards and changes to timelines regularly and at important timepoints. We would love the community’s advice on how to better communicate, our current channels (including blog posts!) may not be effectively reaching the audience for our messages.

Gratitude and Next Steps

The success of this experiment hinges on your willingness to engage with something new and trust us with your research. Whether your paper was selected or not, your contribution to this inaugural year has been meaningful and appreciated. Thank you. 

The conversations that matter most in our field—about ethics, impact, and AI’s future—don’t end with a single conference track. They continue in labs, journals, online discussions, and countless ways our community engages with challenges and opportunities ahead.

Thank you for making the first NeurIPS Position Paper Track possible! We hope you enjoy all of the discussions sparked by these papers at NeurIPS in December!

NeurIPS has grown at an unprecedented pace in recent years, fundamentally reshaping how the conference operates. To provide transparency and share insights into how NeurIPS continues to refine its processes, we—the Program Committee (PC) Chair Team—are publishing this blog post on our experience managing the review process at scale, with support from the Communication Chairs.

NeurIPS has multiple tracks, each with their own chairs and program committees, as well as their own processes and criteria for selecting papers. This year the three tracks that constitute the poster sessions in the main conference  include the Main Program track, Datasets and Benchmarks track and the Position Paper track. While some attempts have been made to align some of these tracks, each track is unique and thus should not adopt the exact same processes. This blog post focuses specifically on the Main Program, but the Datasets and Benchmarks chairs have released a blog post, as have the Position Paper Track chairs.

How the scale of NeurIPS submissions impacts calibration on decision-making

Recruiting SACs, ACs, and Reviewers at Scale

Like most AI conferences, NeurIPS has seen a substantial increase in the number of submissions in the past few years, growing from 9,467 submissions in 2020 to 21,575 in 2025. Critically, this increase in submissions not only is a challenge from the perspective of ensuring that there are enough reviewers, area chairs (ACs), and senior area chairs (SACs) to give papers fair review (thank you to all of the 20,518 reviewers, 1,663 ACs and 199 SACs who work tirelessly to make this happen), but it also introduces effects that make the review process noisier. For example, recruiting has to happen at a larger scale, but this creates a greater possibility that papers might be reviewed and handled by less experienced reviewers/ACs/SACs. For this year, we had to recruit a larger number of ACs from self-nomination forms, so we had a larger number of first-time NeurIPS ACs. Another example is that the NeurIPS community continues to expand in topic breadth over time, with backgrounds ranging from statistical physics to psychology joining the field, and new topics such as  LLMs becoming highly popular as existing fields evolve. This means that the existing pool of reviewers may not be topically qualified to handle these topics – and even experienced reviewers may struggle to cover some topics when the relevant expert communities are still emerging.

Calibration Process
For this reason, PC chairs need to focus more on calibration of decision-making, doing their best to ensure that the reviewing process remains reliable in the presence of this noise. This year, we relied on a range of signals for this – for example, if there is disagreement between reviewers, or between reviewers and the AC, we often look for consensus between the AC and SAC to calibrate. We found it extremely helpful when ACs actively communicated and provided feedback in this process. Many ACs fought for papers they thought were good even if reviewers disagreed, and often we followed their lead, ending in an accept decision for the paper. Conversely, sometimes ACs flagged significant negative issues with evidence after the rebuttal that reviewers did not catch, and we ask SACs to carefully discuss these issues with the ACs to ensure the evidence is convincing. This then led to reject decisions even when papers had high ratings. We understand and we are truly regretful that this might have severely disappointed some authors. However, we trusted the ACs and SACs to be experienced and professional members of the community whose consensus guards the quality of the program. Finally, we note in some cases decisions had to be reverted for reasons outside of what could be disclosed to the AC. This included situations like the 11 unfortunate cases where one of the co-authors are confirmed to be grossly negligent in their reviews under the our new responsible reviewing policies, in line with practices by CVPR and other sister conferences who have adopted similar policies. 

At the same time, we acknowledge that the calibration process is not perfect, and it is impossible for it to be perfect. Calibrating decisions is a constrained optimization process. Everyone involved has limited resources: authors do not have infinite time to comment, and reviewers, ACs, SACs and PCs do not have infinite bandwidth to read and respond. Even if for some individual cases, the decision could be retrospectively improved, we cannot revisit these decisions as if the original constraints driving the decision did not exist. For example, some authors asked why they could not make rebuttals if ACs raised new issues independently of the reviewers, but applying this fairly across reviews would require adding additional weeks to the reviewing timeline, which would derail conference planning due to fixed timeframes.

Upholding Scientific Standards Despite Challenges

Nonetheless, we are optimistic that on balance, NeurIPS decisions were consistent with the standards expected from NeurIPS papers established by prior years. The main track this year received 21,575 valid paper submissions, of which 5,290 were accepted. That is, the main track had an acceptance rate of 24.52%, on par with that of previous years. In general, accepted papers were consistent with reviewer impressions of papers (Figure 1), with the PCs manually reviewing many outliers (but not all, due to time resource constraints.)

Figure 1. A histogram of scores across papers in the main program track, color coded by accept versus reject decisions.

Finally, we want to address the topic of space constraints, since some community members were concerned that papers meeting the NeurIPS standard were rejected for this reason. In principle, space is a constraint we have to be mindful of, as our physical venues can only permit so many papers. We were transparent about this with our SACs and monitored it closely as part of our duty. For the main track, however, as we put our attention on resolving boundary and outlier decisions using the consensus of ACs and SACs during calibration, we reached a comfortably lower number of accepted papers than the capacity of our venue holds. That is, space ultimately played a negligible role compared to scientific merit.

An initiative on responsible reviewing this year

Review processes at NeurIPS are always evolving, and not just because the PC chairs have to keep up with the increased scale of the conference and new contexts – we also want to improve the reviewing process from year-to-year and ensure we help build a sustainable conference process that is responsive to community needs and concerns. For this reason, in addition to our duties directly managing the review process, we often focus on specific areas for improvement from year to year. For this year, we have focused on the topic of responsible reviewing.

Among other initiatives, we announced new policies under the responsible reviewing initiative (see our previous blog post). Towards the goal of improving the quality of reviews, this policy allows PCs and other tracks’ chairs to withhold reviews from authors whose own reviewing obligations were unmet on time. In addition, when the reviewers are grossly negligent during the review process, their co-authored papers need to be desk-rejected. In addition, we paid attention to the issue of collusion rings and fake peer review accounts, which had been hot areas of discussion in the community in previous years. This has led to a lower reliance on bidding, as we could not assume everyone was a good actor in face of evidence some were not. In turn, this led to a greater reliance on the matching system used by OpenReview, which likely introduced different types of noise in reviews this year, increasing our workload in calibration. However, we feel this to be an important trade-off to disrupt these academic integrity challenges.

A call for feedback

Overall, we would like to encourage the community to continue to provide feedback on how to improve the review process at NeurIPS. This feedback is critical for us to understand how the changes we implement from year to year to evolve NeurIPS may affect the community. For example, one new source of noise we observed this year was reviewers increasing their scores just to end back-and-forth discussion with the authors, but being silent or critical in private discussions with the AC, which leads us to believe that increasing levels of author engagement in rebutting their papers is now leading to reviewer fatigue. We encourage the community to raise issues like this with us, so we can factor this into future years’ planning. In particular, we encourage community members to attend the Town Hall, which will be hosted on-site at the conference this year, where they can ask questions and raise comments to the organizers in real-time.

The Datasets and Benchmarks (DB) track, since introduced in 2021, has played a pivotal role in raising the profile of datasets and benchmarks, which is foundational to machine learning, in the NeurIPS community. As this track continues to grow and refine its processes from year to year, we, the DB Chairs, decided to write this blog post to make our process this year transparent. This blog post focuses specifically on the DB track, but the Program Committee chairs have also written a blog post, as have the Position Track chairs.

Raising the standards for dataset and benchmark submissions

As the NeurIPS Datasets and Benchmarks (DB) track continues to mature, its growth in submissions is beginning to stabilize, as is also its establishment within the community. After three years of exponential growth, the track received 1,995 submissions this year. While this is a large number, the increase from the 1,820 submissions received last year is smaller than in previous years (for comparison, there were 987 submissions in 2023), suggesting that the track has begun to stabilize. In light of this maturity, significant steps were taken for the second consecutive year to align its standards and processes with the main track. This strategic alignment aims to ensure that papers involving datasets and benchmarks are held to the same rigorous criteria as main track papers, and maintain the reputation for quality that NeurIPS proceedings papers have. This is also particularly relevant for papers that blur the lines across tracks (most frequently benchmarks and evaluations).

A first step in the direction of alignment with the main track began in 2024, when DB track saw an acceptance rate of 25.3% closely mirroring the main program’s 25.8% rate. For 2025, our goal was to align even more with the main track by working together from recruiting reviewers and ACs, adapting the majority of the main track’s reviewing processes, and finally jointly implementing the responsible reviewing initiative. This initiative is designed to safeguard the quality of reviews, addressing issues of late or low-quality feedback that can hinder the peer-review process. All this was aimed at ensuring a consistent and high level of scrutiny for all submissions. 

Building on the lessons learned from previous years, the DB track chairs have also focused on streamlining the submission and review process for datasets. The objective was to create a more standardized submission process for authors and a more efficient evaluation method for reviewers, and in this to allow for easier access, comparison, and assessment of datasets. We elaborated on these changes in our blog post titled NeurIPS Datasets & Benchmarks: Raising the Bar for Dataset Submissions. We highlighted the evolving nature of AI research, where the distinction between a “dataset paper” and a “main track” paper can be nuanced. To address this, the updated best practices and requirements for 2025 include more stringent criteria for dataset submission, hosting, and reproducibility. These measures are intended to ensure that datasets are not only useful and accessible at the time of publication but also remain so over time, thereby improving their accessibility for the review process.

Many of these enhancements are reflected in the updated Call for Papers for the DB track, which this year directly references the main track’s call for papers while providing specific guidelines for dataset and benchmark submissions. This change sends a clear message about the intent to elevate the quality of the DB track submissions to the well-established standards of the main conference.

DB Track calibration process for ensuring fairness and consistency

A critical component of our commitment to maturing the DB track is the careful calibration of our decision-making process. Like all tracks at NeurIPS, we must ensure that reviewing standards are applied consistently and fairly across all submissions, accounting for the natural variance that exists among reviewers, Area Chairs (ACs), and Senior Area Chairs (SACs). As the Program Committee Chairs have noted for the main track in their blog post, many factors can introduce noise into reviewer scores and feedback. It is our responsibility as the DB track chairs to mitigate this and ensure that every paper receives a fair evaluation.

In line with our primary goal of raising the standards and streamlining processes for the DB track, this year, we collaborated closely with the PC chairs to parallel many of the main track reviewing processes. This marks a significant step forward from previous years, where the DB track largely defined its own standards. While our processes were not identical, as DB papers have unique considerations, we adopted similar protocols to the main track for resolving disagreements, such as supporting ACs and SACs throughout the process of calibration for the final decision making. 

This year, we implemented two significant changes to the review process, hand in hand with the main track. First, we introduced a revised scoring system. Second, we rolled out the responsible reviewing initiative, which focused on safeguarding review quality and timeliness.

The new scoring system may have influenced how reviewers assessed submissions. In particular, we have observed two distinct trends in the DB track during this process:

1. Increased average score: Unlike papers in the main track, which one can expect to be more method and algorithm oriented, it is less common for a dataset submission to be fundamentally “technically incorrect.” 
2. Subjective nature of contributions: Evaluating the merit of a dataset or benchmark can be highly subjective. A dataset that fills a critical gap for a smaller, long-tail research area might be just as valuable or novel as one that targets a well-established “head” problem.

We think that these factors can lead to reviewer scores that skew higher and have a tighter distribution compared to the papers in the main track. As a result, it can be difficult for the ACs to differentiate clearly between two papers that may have the same average score but vastly different merits and trade-offs.

To address this, this year, at the end of the calibration process, which was mirrored from the main track, we implemented an additional step that allowed us to have a nuanced evaluation system, rather than relying solely on numerical scores. Thus, we asked our SACs, based on their discussion with the ACs, to produce a relative ranking of the papers within their stack of papers. The same procedure was also done last year. The main difference is that last year, we did not have a template for the rankings, and SACs were asked to explain their ranking in a live meeting. To be mindful of the increased effort, this year, we structured this interaction using a ranking form. For any paper ranked below the track-wide average score of 4.25, SACs were required to provide a detailed description of its merits and motivation. This combination of relative ranking and qualitative justification provided a much richer signal.

Provide us with feedback

As with all tracks, community feedback is essential for us to continue to improve our processes: please reach out if you have any feedback. This year, we advanced many changes to improve the quality of datasets and benchmarks papers at NeurIPS, under the idea that a NeurIPS DB paper should not just be correct, but also meet standards for impact and scientific relevance. In future years, we would like to better describe these criteria to reviewers/ACs/SACs, and would invite feedback on this, especially.  In particular, we encourage community members to attend our Town Hall at the conference venue, where they can ask questions and provide comments to the organizers in real-time.

The NeurIPS 2025 organizing committee is pleased to announce the invited speakers for this years conference. The lineup includes six researchers representing areas spanning theoretical machine learning, reinforcement learning, AI and society, cognitive science, natural language processing, and deep learning applications.

Kyunghyun Cho

Glen de Vries Professor of Health Statistics, NYU; Executive Director of Frontier Research, Prescient Design, Genentech
Website: kyunghyuncho.me

Cho’s work spans machine learning and natural language processing. He co-developed the Gated Recurrent Unit (GRU) architecture and has contributed to neural machine translation and sequence-to-sequence learning. He is a CIFAR Fellow of Learning in Machines & Brains and received the 2021 Samsung Ho-Am Prize in Engineering. He served as program chair for ICLR 2020, NeurIPS 2022, and ICML 2022.

Yejin Choi

Professor of Computer Science, Stanford University; Dieter Schwarz Foundation  Senior Fellow, Stanford HAI; Distinguished Scientist, NVIDIA
Website: yejinc.github.io

Choi’s research focuses on natural language processing, with emphasis on commonsense reasoning and language understanding. She is a 2022 MacArthur Fellow and was named to Time’s Most Influential People in AI in 2023. She has received multiple Test of Time Awards from ACL and CVPR, and Best Paper Awards at venues including ACL, EMNLP, ICML, and NeurIPS. She previously held positions at the University of Washington and Allen Institute for AI.

Melanie Mitchell

Professor, Santa Fe Institute
Website: melaniemitchell.me

Mitchell’s research areas include AI, cognitive science, and complex systems, with focus on conceptual abstraction and analogy-making in humans and AI systems. She authored “Complexity: A Guided Tour,” which won the 2010 Phi Beta Kappa Science Book Award, and “Artificial Intelligence: A Guide for Thinking Humans,” which was named one of the five best books on AI by both the New York Times and the Wall Street Journal. She received her PhD from the University of Michigan under Douglas Hofstadter, with whom she developed the Copycat cognitive architecture.

Andrew Saxe

Professor of Theoretical Neuroscience & Machine Learning, Gatsby Computational Neuroscience Unit and Sainsbury Wellcome Centre, UCL
Website: saxelab.org

Saxe’s research focuses on mathematical theories of learning in neural networks. He has developed exact solutions for learning dynamics in deep linear networks and studies connections between artificial and biological learning systems. He is a CIFAR Fellow of Learning in Machines & Brains and recipient of the 2019 Wellcome Trust Beit Prize. His work includes theoretical analyses of semantic development and the dynamics of representation learning.

Richard Sutton

Research Scientist, Keen Technologies; Professor, University of Alberta; Chief Scientific Advisor, Amii Chief Scientific Officer, ExperienceFlow.ai.
Website: incompleteideas.net

Sutton co-developed temporal difference learning and policy gradient methods in reinforcement learning. He received the 2024 Turing Award with Andrew Barto for foundational contributions to reinforcement learning. He is co-author of the textbook “Reinforcement Learning: An Introduction” and is a Fellow of the Royal Society and the Royal Society of Canada. His research focuses on computational principles underlying learning and decision-making.

Zeynep Tufekci

Henry G. Bryant Professor of Sociology and Public Affairs, Princeton University; New York Times Columnist
Website: zeynep.me

Tufekci examines the interplay of science, technology and society through a sociological framework and complex systems lens, focusing especially on digital, computational, and artificial intelligence technologies. She was a 2022 Pulitzer Prize finalist for commentary on the COVID-19 pandemic. Her book “Twitter and Tear Gas: The Power and Fragility of Networked Protest” examines the dynamics of social movements in the digital age. She is also faculty associate at the Berkman Klein Center for Internet & Society at Harvard University.

We are excited to be running a Position Paper Track for the first time in NeurIPS history. This track gives our community an important forum for raising and discussing essential topics in machine learning within and around our community of experts.

This also means that the Position Paper Track operates a little differently from the Technical Research Track, and we wanted to take a moment to elaborate on some of the differences that we are testing with the Position Paper Track this year! 

The biggest difference between the position papers and the other tracks is that we will not be having a traditional rebuttal period this year. Instead, we will have a private opportunity for authors to flag concerns with the integrity of any reviews received, but we are asking that discussion of the substance of papers take place in other community-facing forums. We are currently working to create these spaces. This is in an effort to encourage public discussion around the topics tackled in the papers. 

Rebuttals are usually used to clarify some point of misunderstanding the authors feel may exist in a particular review. Generally, this is around the details of the experiment being conducted. As in position papers, we are not evaluating experiments, but trying to create an open discussion around the points being posed. Rather than using private review dialogues, we wanted to encourage any back and forth to take place in public either for instance in OpenReview or in other fora of the authors’ choosing.  

We are dedicated to a fair and equitable review process. Authors will have the opportunity to voice any concerns they have about the quality of the reviews they received through a survey that will become open to them through Open Review after reviews are sent out to authors. 

This is part of a larger effort to stimulate conversation in the community through the position paper track. Stay tuned for more on our other efforts. We look forward to continuing to innovate on how papers will be presented to promote broader community dialogue. 

As a reminder, this is the timeline for the Position Paper Track:. 

• Assign ACs papers July 2nd
• Assign papers to Reviewers July 3rd
• Reviewer & AC cohort conversation: August 8th–August 28th, 2025
• Surveys become available to authors: Week of August 10th 
• Deadline for meta reviews: August 28th, 2025
• Author notification: Sept 18, 2025

We want to thank everyone for participating in the first ever Position Paper Track at NeurIPS!

We are excited to announce an expansion of the reach and accessibility of the conference! 

For the first time, in 2025, NeurIPS will host a second official physical location for the conference. The conference will be held at the San Diego Convention Center and in Mexico City. 

This pilot program of  a second official location will allow us  to continue supporting exchange and communication across the field of AI research while responding to skyrocketing attendance and difficulties in obtaining travel visas some attendees have experienced in the past few years when only one location was available. 

The Mexico City location will host approximately five hundred attendees. Programming for the main conference will mirror that which is taking place in San Diego. NeurIPS hopes to support programming for the workshops and tutorials that have been accepted for the 2025 program to also appear in Mexico City.

Authors whose work has been accepted to the conference but are unable to travel to San Diego will be given priority for attendance in Mexico City. We hope that by providing two options for in person attendance we can relieve some of the burden of travel while promoting scientific exchange. 

As we get closer to December, more details about programming, logistics, and attendance at all NeurIPS locations will be available. 

We hope to see you at the conference!