NeurIPS 2021 Paper Checklist Guidelines

Abstraction: NeurIPS submission checklist for responsible ML research transparency standards

Key points:

• Checklist covers eight domains: main claims accuracy, ethics review compliance, negative societal impacts, limitations disclosure, theoretical proofs, experiment reproducibility, asset licensing and consent, and human subject research protections
• Answering "no" or "n/a" with justification is fully acceptable; the checklist is not grounds for rejection but shapes reviewer evaluation
• Societal impact section requires discussing potential harms (disinformation, surveillance, fairness, privacy, adversarial attacks) and mitigation strategies (gated release, monitoring, efficiency)
• Compute section encourages reporting GPU type, compute per run, total compute, and CO2 emissions using tools like ML CO2 Impact Calculator, CodeCarbon, or experiment-impact-tracker
• Reproducibility requires exact commands, environment specs, training details (data splits, hyperparameters), and error bars or confidence intervals on main results
• Checklist designed to be consulted early during research to positively shape methodology, not as a post-hoc afterthought

Connections: Neurips · Research Ethics · Reproducibility · Responsible AI

Source: https://neurips.cc/Conferences/2021/PaperInformation/PaperChecklist