Methods with rigorous evaluation
New models evaluated beyond in-distribution error on a static test case.
Machine Learning/CORE Unranked
AI4PowerGrids 2026
AI Foundations for Power Grids @ NeurIPS 2026
AI4PowerGrids 2026 is a NeurIPS workshop focused on advancing machine learning for power grid applications, emphasizing realistic evaluation under physics constraints, structural shifts, and real-time operational challenges. It aims to bridge the gap between ML research and power systems by promoting rigorous benchmarks, failure mode analysis, and system-level validation.
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Benchmarks, datasets, and evaluation protocols
Contributions whose primary artifact is how we measure.
Position and empirical-evaluation papers
Systematic studies, audits, or arguments about what good evaluation should look like.
Negative results and failure modes
Short papers documenting where learned components break, especially under structural or distributional shift.
Research areas in scope
01
Themes
Open, realistic datasets as a first-class community deliverable, beyond legacy IEEE test casesFeasibility as a metric – what it means for a stochastic model to 'respect the physics', beyond aggregate errorStructural, not just sample, shift – real grids change topology hourly and generation mix yearlyFoundation-model claims for grids and the criteria by which to evaluate themComponents vs. systems – a 99% accurate surrogate can destabilise the loop it sits insideFrom eyes-on to eyes-off – offline, advisory, and narrow-autonomy stages each demand different evidence
Policies worth checking twice
- Submissions must be double-blind: remove author names, affiliations, acknowledgments, grant identifiers, repository usernames, and personal/institutional project pages.
- Cite your own prior work in the third person.
- All code, data, models, or demonstrations shared during review must be hosted at an anonymous link or repository.
- Submissions must include a short domain checklist addressing physics feasibility, out-of-distribution evaluation, failure modes, system-level effect, and data/code availability.
- Do not include the standard NeurIPS Paper Checklist; only the domain checklist is required.
- Length limit: up to 4 pages of main content, unlimited references; domain checklist does not count toward the page limit.
- Submissions must be in English and uploaded as a single PDF via OpenReview using NeurIPS 2026 LaTeX style files.
- Work already published or accepted at NeurIPS, ICML, ICLR, AAAI, or comparable ML venues is not eligible.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026