NeurIPS 2025 Workshop on
Efficient Reasoning
Dec 6 or 7, 2025
Contact: er.neurips2025@gmail.com




News

  • Call for reviewers: We are actively looking for reviewers to join the program committee for the workshop. We encourage all interested researchers to apply, especially those from underrepresented groups. Prior reviewer experience is a nice-to-have, but not required. Interest and familiarity with subject matters related to the workshop is required. If you are interested, please fill out the application form to join us.
  • Call for papers: Submission portal will open at OpenReview soon

About

Recent progress in large reasoning models (LRMs), like OpenAI o1 and Deepseek R1, has been pivotal for tackling complex applications, from mathematical and code reasoning to advanced symbolic and agentic planning. Their success often relies on test-time scaling, which involves increasing the generation length or depth. However, these approaches incur significant efficiency bottlenecks during training and inference. To overcome these limitations, further advancements are needed in data, algorithms, and systems applicable across various domains, as exemplified by work such as s1, Z1, and verl. The proposed workshop will bring together researchers and practitioners to rethink efficient reasoning under tight compute, memory, latency, throughput, and cost budgets, with the goal of translating theoretical breakthroughs into practical, deployable solutions.

Key Problems We Aim to Address

Dataset Curation
Design, create, and maintain high‑quality datasets for training and evaluating LRMs on challenging reasoning tasks under resource constraints. Contributions include long-context reasoning, symbolic planning, multi-hop deduction, and real-time decision-making on diverse domains.
Algorithmic Innovation
Develop efficient training algorithms for supervised fine-tuning or reinforcement learning, and efficient inference methods like pruning, compression, progressive generation, and search‑based strategies, reducing time/space complexity without eroding accuracy or faithfulness.
System Deployment
Implement efficient RL training systems and long-CoT inference engines to support current LRMs, while unlocking new reasoning potential on commodity GPUs. Techniques may include dynamic KV‑cache placement, quantized graph execution, and on‑device knowledge distillation
Application
Adapt LRMs to enable real-time reasoning in resource-constrained scenarios, like clinical decision-making, robotics, autonomous driving, developing-world health, and on-orbit autonomy.

Call for Papers

The 1st Workshop on Efficient Reasoning at NeurIPS 2025 invites submissions from researchers focused on the reasoning models. Additionally, we welcome contributions from researchers in the natural sciences (e.g., physics, chemistry, biology) and social sciences (e.g., pedagogy, sociology) to provide attendees with a more comprehensive perspective for their reasoning tasks.

Key Dates

  • Paper Submission Open: August 1, 2025
  • Paper Submission Deadline: August 22, 2025 (AoE)
  • Paper Notification Deadline: September 22, 2025 (AoE)
  • Camera-ready Version Deadline: October 7, 2025 (AoE)

Deadlines are strict and will not be extended under any circumstances. All deadlines follow the Anywhere on Earth (AoE) timezone.

Submission Site

Submit papers through the Workshop Submission Portal on OpenReview.

Scope

We welcome contributions across a broad spectrum of topics, including but not limited to:

  • Pipelines for creating high-quality training data with resource constraints or deployment environments.
  • Benchmark methodologies for assessing the efficiency and efficacy of LRMs in real-world settings
  • Innovations in techniques for training LRMs with a better trade-off between efficiency and performance
  • Approaches for accelerating LRM inference through the design of algorithms and systems
  • Advancements in improving the reasoning and planning abilities of LRMs in diverse tasks
  • Theoretical analysis of the time and space complexity of LRMs on synthetic and realistic tasks.
  • Empirical investigations into the practical efficiency (i.e., latency and throughput) of LRMs
  • Implementations to efficiently enable large-scale RL training systems or on-device inference engines
  • Strategies for overcoming practical limitations (e.g., memory, time, data) of LRMs
  • In-depth discussions exploring the efficient deployments and applications of LRMs

Submission Guidelines

Format:  All submissions must be a single PDF file. We welcome high-quality original papers up to 4 pages. References and appendices are not included in the page limit, but the main text must be self-contained. Reviewers are not required to read beyond the main text

Style file:   You must format your submission using the NeurIPS 2025 LaTeX style file. For your convenience, we modified the main conference style file to refer to our workshop: NeurIPS 2025 Style Files. Please include the references and supplementary materials in the same PDF. The maximum file size for submissions is 50MB. Submissions that violate the NeurIPS style (e.g., by decreasing margins or font sizes) or page limits may be rejected without further review.

Dual-submission and non-archival policy:  We welcome ongoing and unpublished work. We will also accept papers that are under review at the time of submission, or that have been recently accepted, provided they do not breach any dual-submission or anonymity policies of those venues. The workshop is a non-archival venue and will not have official proceedings. Workshop submissions can be subsequently or concurrently submitted to other venues.

Visibility:   Submissions and reviews will not be public. Only accepted papers will be made public.

Double-blind reviewing:   All submissions must be anonymized and may not contain any identifying information that may violate the double-blind reviewing policy. This policy applies to any supplementary or linked material as well, including code. If you are including links to any external material, it is your responsibility to guarantee anonymous browsing. Please do not include acknowledgements at submission time. If you need to cite one of your own papers, you should do so with adequate anonymization to preserve double-blind reviewing. Any papers found to be violating this policy will be rejected.

Contact:   For any questions, please contact us at er.neurips2025@gmail.com or EfficientReasoning2025@googlegroups.com


Schedule

This is the tentative schedule of the workshop. All slots are provided in local time.

Morning Session

08:50 - 09:00 Introduction and opening remarks
09:00 - 09:30 Invited Talk 1
09:30 - 10:00 Invited Talk 2
10:00 - 10:15 Contributed Talk 1
10:15 - 11:15 Poster Session 1
11:15 - 11:45 Invited Talk 3
11:45 - 12:15 Invited Talk 4
12:15 - 13:30 Break

Afternoon Session

13:30 - 14:00 Invited Talk 5
14:00 - 14:30 Invited Talk 6
14:30 - 14:45 Contributed Talk 2
14:45 - 15:45 Poster Session 2
15:45 - 16:15 Invited Talk 7
16:15 - 16:30 Contributed Talk 3
16:30 - 17:00 Invited Talk 8
17:00 - 18:00 Panel discussion

Invited Speakers

This is the tentative invited speakers. More speakers are invited.




Beidi Chen

Carnegie Mellon University

Arman Cohan

Yale University

Jonas Geiping

ELLIS Institute

Haibin Lin

ByteDance Seed

Niklas Muennighoff

Stanford

Hongyu Ren

Meta

Yuandong Tian

Meta

Hao Zhang

UCSD

Workshop Organizers




Cheng Luo

California Institute of Technology

Xinyu Yang

Carnegie Mellon University

Simran Arora

Stanford

Weijia Shi

University of Washington

Hanshi Sun

ByteDance Seed

Songlin Yang

MIT CSAIL

Luca Zancato

Amazon

Jiawei Zhao

Meta