Call for Contributions

Call for Extended Abstracts

We invite submissions of extended abstracts (2–4 pages, excluding references) that advance the field of constrained learning (deadline: ~~Aug 21, 2025~~ Aug 28, 2025 (AOE). Final versions will be posted on the Papers tab but will not be part of formal proceedings. Accepted abstracts will be presented at the workshop during two poster sessions, with selected contributions invited to give a short talk (see Contributed Talks for details).

Scope: We welcome submissions on constrained optimization and its application to enforcing properties on machine learning models, with a focus on work that develops or leverages tailored constrained optimization techniques. In general, this excludes approaches that impose properties via penalties or regularization—that is, by augmenting the loss with a linear penalty term and minimizing the resulting unconstrained objective. (See this Position Paper for a discussion of why such methods are typically not principled in the context of constrained optimization.)

We particularly encourage contributions in the following areas:

Algorithms: Fundamental algorithms for constrained optimization, especially those suited to constrained deep learning problems—i.e., differentiable, large-scale, non-convex, and stochastic settings. We welcome approaches that offer favorable efficiency-complexity tradeoffs and hyperparameter robustness in real-world applications.
Constraint Regularization: Techniques that improve constraint satisfaction on unseen data.
Learning Theory: Statistical learning theory for constrained problems, including results on constraint generalization, convergence, and stability.
Continuous Relaxations: Techniques for solving discrete constrained problems via continuous relaxations.
Learning to Optimize approaches applied to solving constrained learning problems.
Applications: Practical applications of constrained optimization in deep learning. We particularly encourage submissions that highlight the crucial role of constraints in safety-critical domains (e.g., autonomous vehicles, medical diagnosis) or trustworthy AI (e.g., fairness, robustness, interpretability). We also welcome applications that expose methodological gaps and motivate new research directions.
Software: Libraries and tools that support constrained deep learning workflows.

Submissions on both constrained minimization and constrained game formulations are welcome.

Submission Guidelines

Please format your submission using the modified NeurIPS style file available here: coml_styles.zip. (This is the NeurIPS 2025 style file with an updated footer indicating submission or acceptance at this workshop rather than the main conference).
Submissions should be 2–4 pages long, excluding references and appendices.
Appendices are accepted but discouraged; the reviewers will not be required to read the appendices.
Submissions will be managed through the OpenReview portal.

Please note the following requirements:

Submissions must be original work in progress not previously published in peer-reviewed conferences or journals (see Dual Submissions and Previously Published Work below for details).
At least one author is expected to attend the workshop in person, and we encourage all authors to participate.
Submissions are not required to include a checklist.

Submissions will be reviewed by a panel of 3 reviewers. The review process will be double-blind, and authors should ensure that their submissions do not reveal their identities.

To better understand the criteria reviewers will use when evaluating your submission, please refer to the Reviewer Guidelines.

Dual Submissions and Previously Published Work

Following the NeurIPS 2025 workshop policies, submissions must be original work in progress not previously published in peer-reviewed conferences or journals. Submissions found to have been previously published will be desk-rejected. Note that work presented at prior non-archival workshops is permitted.

Dual submissions are allowed, provided the other venue is either (i) non-archival (e.g., another workshop), or (ii) archival with a publication date after the COML notification date (September 22, 2025). If you are submitting to another venue, please ensure that your COML submission complies with that venue’s dual submission policy.

Contributed Talks

Four extended abstracts will be selected for a short contributed talk at the workshop. Selection will be based on the quality and relevance of the abstract to the workshop themes. To promote a diverse set of contributions, we aim to select one talk from each of the following categories:

Best fundamental (theoretical or algorithmic) contribution,
Best application,
Best overall contribution,
Best negative results contribution.

However, the final selection may deviate from these categories depending on the quality of submissions.

Each talk will be 10 minutes long, followed by a 5-minute Q&A session, and must be delivered in person. Oral presenters will be notified by Sep 29, 2025.

Important Dates

As per OpenReview guidelines, please ensure that you create an account at least two weeks before the submission deadline to avoid any delays.

Submission deadline: ~~Aug 21, 2025~~ Aug 28, 2025 (AOE)
Author notification: Sep 22, 2025
Oral selection notification: Sep 29, 2025
Camera-ready version: Oct 31, 2025 (AOE)

Please note there will be no discussion period; reviews and decisions will be final.

Code of Conduct

COML authors are expected to adhere to the NeurIPS 2025 Code of Conduct and the NeurIPS 2025 LLM Policy for Authors. In particular, while LLMs may be used to assist in writing submissions, authors remain fully responsible for the content, which must not be entirely generated by an LLM.

NeurIPS 2025 Workshop on Constrained Optimization for Machine Learning