NeurIPS 2025 Workshop on Constrained Optimization for Machine Learning

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 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).

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.
  • 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. Work presented at prior non-archival workshops is allowed.
  • 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.
  • We follow NeurIPS policies, including their guidelines on the use of large language models.

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 negative results contribution,
  • Best overall contribution,
  • Best application,
  • Best fundamental (theoretical or algorithmic) 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 (AOE)
  • Author notification: Sep 22, 2025
  • Oral selection notification: Sep 29, 2025
  • Camera-ready version: Oct 31, 2025 (AOE)

Please note that we are unable to provide financial support for travel or accommodation. If you require assistance to attend the workshop, we encourage you to check the NeurIPS financial assistance page.

Questions?

Contact us at constrainedml@gmail.com.