Short papers
Up to 5 pages, reporting new research or arguing a position; references and appendices do not count toward the page limit.
Machine Learning/CORE Unranked
InterpScience 2026
Interpretability as a Science: NeurIPS 2026 Workshop
InterpScience 2026 is a NeurIPS workshop focused on establishing interpretability of large language models as a rigorous empirical science. It seeks to define standards for measurement, causality, and falsifiability in interpretability research, drawing insights from neuroscience, statistics, and causal representation learning through interactive breakout sessions and discussions.
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Long papers
Up to 9 pages, reporting new research or arguing a position; references and appendices do not count toward the page limit.
Research areas in scope
01
Topics of interest
What it means to understand a model, and criteria for genuine explanationFormal and mathematical frameworks for interpretabilityCausal and interventional methods for grounding interpretability claimsMeasurement validity, identifiability, and evaluation designFalsifiability and experimental designs that distinguish mechanisms from artifactsTransferable lessons from neuroscience, cognitive science, statistics, econometrics, and related fieldsPathways from interpretability findings to replicable scienceInterpretability work on architectures other than LLMs
Policies worth checking twice
- All submissions are double-blind: no identifying information in main text, figures, or supplementary materials.
- References and appendices do not count toward page limits.
- All camera-ready papers must be in NeurIPS format.
- Work must not have been previously published at another machine learning or related conference, including NeurIPS 2026 main conference.
- Dual submission to NeurIPS main conference is allowed, but accepted papers will not receive a poster or presentation slot at the workshop.
- Dual submission to other NeurIPS 2026 workshops is not allowed.
- Submissions must be made via OpenReview.
- Authors are fully responsible for verifying correctness and originality of content generated with AI/LLMs.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026