Constrained Learning for Scientific Modelling (Paris NeurIPS pre-event) [N]

CALL FOR CONTRIBUTIONS

CL4SM session at ELLIs UnConference, a NeurIPS Paris pre-event: https://badulion.github.io/cl4sm\_2026/

Machine learning models of scientific and engineering systems can fit the data well and still be physically wrong: they create or destroy mass and energy, produce negative concentrations, violate boundary conditions, or drift into unphysical regimes. CL4SM promotes discussion on machine learning models that respect conservation laws, symmetries, stability, and other prior knowledge.

TOPICS include, but are not limited to:
- Soft constraint enforcement (penalty and augmented-loss formulations)
- Hard enforcement by construction (structure-preserving architectures)
- Projection, correction, and differentiable optimisation layers
- Constraints from governing equations, symmetries, boundary conditions
- Stability and safety of learnt dynamical models
- Probabilistic constraints and physically consistent uncertainty
- Trade-offs between guarantees, accuracy, and cost
- Applications in climate, fluids, engineering, energy, and industry

WE INVITE 2-page submissions (including references) in two categories:
1. Ongoing work: new methods, theory, benchmarks, or applications, including early-stage ideas and negative results (talk and poster)
2. Tutorials, demos, software libraries, and datasets (lightning pitch and demo)

We also welcome RECENTLY PUBLISHED PAPERS from top-tier journal and conference venues such as JMLR, NeurIPS, ICML, ICLR, AISTATS, IEEE TPAMI, Machine Learning, etc. : state the original venue and upload the published PDF (talk and poster).

Submissions 2-page submissions use the PMLR template and are reviewed single-blind by the
Program Committee. The session is non-archival.

IMPORTANT DATES
- Submission deadline: October 30, 2026 (23:59 AoE)
- Notification: November 12, 2026
- Final programme: November 27, 2026
- Session: December 8, 2026, Palais des Congrès de Paris (just before NeurIPS 2026)