InterpScience

NeurIPS 2026 · Sydney

Call for Papers

We invite submissions that engage with the scientific foundations of interpretability—measurement, causality, falsifiability, and lessons from adjacent sciences.

Important dates

All deadlines are Anywhere on Earth (AoE).

Paper submission deadline
September 01, 2026
Reviewing deadline
September 18, 2026
Author notification
September 29, 2026
Camera-ready deadline
November 15, 2026
Workshop
December 11 or 12, 2026

Topics of interest

We welcome work on topics including, but not limited to:

  • What it means to understand a model, and criteria for genuine explanation
  • Formal and mathematical frameworks for interpretability
  • Causal and interventional methods for grounding interpretability claims
  • Measurement validity, identifiability, and evaluation design
  • Falsifiability and experimental designs that distinguish mechanisms from artifacts
  • Transferable lessons from neuroscience, cognitive science, statistics, econometrics, and related fields
  • Pathways from interpretability findings to replicable science

Interpretability work on architectures other than LLMs is in scope — see the FAQ.

Submission guidelines

  • We invite submissions of short (up to 5 pages) and long papers (up to 9 pages), reporting new research or arguing a position. References and appendices do not count toward those limits; see the FAQ.
  • We accept both ICLR and NeurIPS formats for submissions, but all camera ready papers must be in NeurIPS format. Find the style template at https://neurips.cc/Conferences/2026/CallForPapers.
  • All submissions are double-blind: authors should ensure the submitted draft contains no identifying information, such as author names, affiliations, acknowledgements, or links that reveal identity, in the main text, figures, or supplementary materials.
  • Accepted papers will be non-archival and presented as posters, with a subset selected for contributed talks.
  • All submissions will require at least one reciprocal reviewer, unless no authors are qualified to serve as a reviewer or program committee member, in which case an exception can be requested. The FAQ explains how to request one.
  • Each reviewer will be asked to review up to 4 papers, tentatively between Sept 03-17, 2026. Please sign up at https://forms.gle/DCUMr9WMWwn3pN6FA to express interest in joining the program committee.
  • Per NeurIPS's policy on the use of agents and LLMs, authors are fully responsible for verifying the correctness and originality of all paper content, including text, figures, and references generated with the help of such tools. Submissions containing fabricated or non-existent citations (e.g., hallucinated references) will be desk rejected without review.

Conceptual and perspective papers count as position papers here — see the FAQ.

Dual submission policy

    As per NeurIPS guidelines:

  • We do not accept work that has previously been published at another conference on machine learning or related fields.
  • Work accepted to the main conference cannot also appear at the workshop. Since the NeurIPS acceptance decisions are not released until after the workshop submission deadline, we allow submissions that are concurrently under review at NeurIPS, with the understanding that these will not receive a poster or presentation slot at the workshop if accepted to the main conference.

    • Additional workshop policy:

  • To promote diversity of papers across NeurIPS workshops, we do not allow submissions currently under review at another workshop (including other workshops co-located with NeurIPS 2026).

Submit on OpenReview.