Icml Supplementary

ICML Supplementary

Use this to place material correctly. ICML 2026 allows references and unlimited appendices in the same PDF, plus supplementary manuscripts and code/data, but critical evaluation material should be in the main paper because reviewers may choose whether to inspect supplements.

Placement map

• Main 8-page body: contribution, method, essential assumptions, central theory, headline experiments, limitations, impact statement placement, and evidence needed for judgment.
• Same-PDF appendices: proofs, extended algorithms, expanded ablations, extra qualitative examples, dataset details, and analysis that supports but does not define the claim.
• Supplementary manuscript: anonymized concurrent submissions or unusually long supporting text that cannot fit cleanly in the paper PDF.
• Code/data supplement: anonymized code, data, scripts, environment, and README.
• Camera-ready public artifact: final code/data repository or archive. ICML 2026 does not provide a final camera-ready supplementary-material upload.

Concurrent submission handling

If an ICML submission cites another concurrent ICML submission by overlapping authors, include the anonymized PDF in supplementary material and explain the relationship in the main paper.

Placement decision table

ICML reviewers may skip supplements, so misplacement is a real acceptance risk, not a formatting nit. Decide placement by whether a reviewer can judge the claim without the material.

Worked vignette: optimizer paper placement

For an adaptive-step optimizer with a non-convex convergence theorem plus deep-learning benchmarks, the convergence theorem statement, one decisive benchmark, and the component ablation stay in the 8-page body. The full proof, the step-size sensitivity sweep, and per-seed tables go to the same-PDF appendix. The training scripts and environment file go to the code/data supplement. Because accepted ICML papers can publish original supplementary material, scrub the code supplement of identity leaks and unreleasable data before upload. A common reject pattern is burying the only fair tuned baseline in the appendix where a heavily loaded reviewer never looks. When the appendix is unlimited, the temptation is to defer hard results; resist it, because at ICML the decisive evidence has to sit in the main body where the soundness judgment is actually formed.

Output format

[Main body] <items that must stay>
[Appendix PDF] <items>
[Supplementary manuscript] <items>
[Code/data supplement] <items>
[Camera-ready public artifact] <items>

Source: brycewang-stanford/Awesome-Journal-Skills → ICML-Skills/skills/icml-supplementary/SKILL.md