Submission guidelines
Tentative requirements
The submission formats and requirements below are provisional. We may expand them to welcome additional forms of work. Any changes will be announced on this page.
All submissions must be made through the OpenReview submission portal ↗ and remain private during review. We welcome methodological, empirical, theoretical, and position work on using interpretability to uncover new knowledge.
Failure cases and negative results are welcome. We encourage careful studies of unsuccessful methods, misleading interpretations, failed validation, and practical limitations that help establish when interpretability can or cannot support reliable discovery.
Formatting
- Write the submission in English and upload it as a single PDF.
- Use the NeurIPS 2026 workshop LaTeX template.
- References and appendices are excluded from the five-page main-text limit.
- The main text must be self-contained. Reviewers are not required to read appendices.
Plan ahead
If you do not have an institutional email, OpenReview account approval can take up to two weeks. Please create your account well before the submission deadline.
Presentation and camera-ready
- Accepted papers will be presented as posters, with a subset selected for oral or spotlight talks.
- Accepted papers may use one additional main-text page in the camera-ready version, allowing up to six main-text pages to integrate reviewer feedback.
- Authors are encouraged, but not required, to attend the workshop in person.
Eligibility and prior work
- Work previously submitted to or accepted at a non-archival venue, including workshops, is welcome.
- We do not accept work that has already been accepted at an archival venue, with the exception of NeurIPS 2026 papers submitted through the fast-track process below.
- Submissions undergoing peer review at another venue, including ICLR or NeurIPS, at the paper submission deadline are welcome.
NeurIPS 2026 fast track
We welcome papers that have already been accepted to NeurIPS 2026. Authors may request fast-track consideration and will be asked to provide evidence of the prior reviews and acceptance. Fast-track submissions remain subject to additional review, including workshop theme fit, after the regular reviewing process is complete.
Anonymity and reproducibility
Double-blind review
The reviewing process is double-blind. Authors are responsible for ensuring that the manuscript and supplementary materials contain no identifying details.
- Remove author names, affiliations, acknowledgments, and other identifying details. Cite prior work in the third person.
- Before submission, search the manuscript for the names, GitHub usernames, and Hugging Face usernames of all core contributors.
Code, data, models, and demos
Authors are strongly encouraged to open source relevant code, models, prompts, data, and interactive demos. Reviewers will be specifically asked to consider reproducibility and the availability of code and data.
- For GitHub repositories, we recommend anonymous.4open.science ↗.
- For larger files such as model weights and datasets, we recommend using an anonymous Hugging Face account.
- For interactive demos, we recommend an anonymous website or Streamlit app.
Review and publication policies
Double-blind review
Submissions must be fully anonymized. Authors are responsible for removing all identifying details.
Non-archival
Prior non-archival work and work under review elsewhere are welcome. Workshop acceptance does not prevent later archival publication.
Responsible-use statement
Every submission must include a short statement covering potential societal impacts and suggested mitigations. A missing statement is grounds for desk rejection.
Camera-ready revision
Accepted papers receive one additional main-text page, for a maximum of six pages, to incorporate reviewer feedback.
Contact
Questions may be sent to interp4discovery@gmail.com.