ICML Position Paper: Structure & Submission Guide

An ICML position paper is a scholarly manuscript submitted to the dedicated Position Paper Track of the International Conference on Machine Learning (ICML) that argues a viewpoint, perspective, or normative stance on what the machine learning research community should do. In contrast to main conference research tracks that document completed technical developments or novel empirical algorithms within machine learning, a position paper adopts a meta-level perspective across the discipline. Because conference proceedings serve as primary publication venues in computer science, understanding authorship credit in computer science conference papers helps research teams establish proper attribution when authoring agenda-setting perspectives.

Purpose of the section

The primary purpose of the ICML Position Paper Track is to stimulate constructive, civil discussion on critical, timely topics affecting machine learning research, governance, and deployment. Rather than rewarding incremental benchmark gains, the track evaluates submissions on whether they present a compelling, non-obvious argument that warrants broad exposure across the community. Submissions address high-stakes issues including training data legality, copyright, intellectual property, privacy guarantees, open-source versus proprietary models, regulatory standards, research ethics, and institutional evaluation metrics.

Reviewers evaluate position papers using distinct criteria from the main research track. Evaluation focuses on seven core dimensions: Position (clear advocacy of a normative stance), Support (reasoning and evidence), Significance (timeliness, scope, and impact), Discussion potential (capacity to provoke productive debate regardless of whether reviewers personally agree), Alternative views (fair treatment of counter-arguments), Communication quality (clarity and structure), and Context (thorough grounding in relevant literature).

Structure

Submissions to the ICML Position Paper Track must adhere to strict structural and formatting conventions specified in the conference call for papers:

Title with Mandatory Prefix: The title must assert a clear position and explicitly start with the prefix “Position:” (for example, “Position: Stop Research on Psychic Properties of Machine Learning”). Generic or descriptive titles that merely state a topic (such as “Position: A Perspective on Model Interpretability”) fail this requirement.
Abstract with Explicit Thesis: The abstract must identify the manuscript as a position paper and clearly state its central thesis, using the standard phrasing: “This position paper argues that [statement of position].”
Introduction with Bold Thesis Statement: The introduction must contextualize the issue within the broader ML field and explicitly state the core position using bold text.
Scholarly Argumentation and Evidence: The body develops logical reasoning, conceptual frameworks, or empirical evidence supporting the position, adhering to rigorous scholarship standards and formal literature citations.
Mandatory Alternative Views Section: Authors must include a dedicated “Alternative Views” section in the main body of the paper (not in an appendix). This section must articulate and address viable, non-strawman opposing perspectives.
Actionable Recommendations and Call to Action: The discussion highlights concrete policy reforms, methodological changes, or research trajectories for researchers, funding agencies, or conference committees.
References: Rigorous attribution of prior work adhering to standard ICML citation guidelines, with no page limits imposed on bibliography entries.

A short annotated example

The following example contrasts a compliant position paper framing with a non-compliant technical submission:

Compliant Framing

Title: Position: Machine Learning Benchmarks Must Report Environmental Compute Footprints Alongside Accuracy

Abstract: This position paper argues that standard ML evaluation benchmarks inadvertently disincentivize efficient architectures by optimizing solely for top-1 accuracy without accounting for training compute. We propose standardized compute disclosures for all benchmark leaderboards.

Introduction Excerpt: Current evaluation paradigms in deep learning suffer from an unaddressed structural bias toward parameter scaling. We argue that conference evaluation committees must mandate verified reporting of kilowatt-hour compute consumption alongside predictive metrics for all official benchmark submissions.

Alternative Views Excerpt: Opponents argue that tracking hardware-specific wattage penalizes labs without access to specialized power-metering infrastructure. However, software-based carbon tracking tools offer acceptable standardized proxies that mitigate hardware monitoring disparities without imposing undue overhead.

Non-Compliant Framing

Title: Position: An Analysis of Compute-Efficient Vision Transformers

Defect: The title lacks an explicit position statement and describes an empirical evaluation rather than an actionable stance. Submissions describing new technical results without advocating a position are redirected or rejected from the position paper track.

Common mistakes

Omission of Mandatory Title Prefix: Failing to lead the title with “Position:” or failing to state an assertive position in the title string.
Placing Alternative Views in an Appendix: ICML guidelines require the Alternative Views discussion to reside within the 8-page main body. Relegating opposing arguments to supplementary material is an explicit structural violation.
Constructing Strawman Opponents: Presenting trivial or weak counterarguments rather than addressing substantive, well-grounded alternative viewpoints held by credible researchers.
Disguising Incomplete Technical Work as a Position: Attempting to salvage negative or unvalidated empirical findings by wrapping them in opinion prose without advancing a normative thesis.
Omitting Bold Text for the Core Thesis: Overlooking the formatting rule requiring the primary position statement to appear in bold typography within the introduction.
Assuming Reciprocal Reviewing Applies: Unlike main track submissions, ICML position papers are exempt from reciprocal reviewing obligations, though accepted manuscripts must provide a lay summary for public communication.

Checklist

Does the manuscript title begin with “Position:” and assert an explicit position?
Does the abstract state “This position paper argues that…” followed by the thesis?
Is the core thesis highlighted in bold text inside the introduction?
Is there a dedicated “Alternative Views” section located within the 8-page main text?
Does the alternative views discussion engage genuine, credible counter-arguments rather than strawmen?
Does the paper adhere strictly to the 8-page limit for main body content (excluding references)?
Have all author identities and institutional affiliations been fully anonymized for double-blind peer review?
Has the correct OpenReview submission portal (“Position Paper Track”) been selected?

Discipline differences

While ICML pioneered a formal track structure, the Conference on Neural Information Processing Systems (NeurIPS) also operates a dedicated Position Paper Track. Researchers authoring perspectives across computer science should understand track differences:

Specification Feature	ICML Position Paper Track	NeurIPS Position Paper Track
Main Body Page Limit	8 pages (excluding references)	9 pages (excluding references)
Title Format Requirement	Mandatory “Position:” prefix; must state position	Must state position clearly; prefix optional
Thesis Formatting in Intro	Must state position using bold text	Must state position using bold text
Alternative Views Section	Mandatory in main text body (never in appendix)	Expected in main body; addresses counter-views
Paper Checklist Requirement	Follows main CFP ethics guidelines	Standard NeurIPS Paper Checklist is waived
Reciprocal Reviewing	Waived for position paper authors	Follows cohort reviewer assignments

Authors navigating AI ethics and publication standards should also review publisher policies such as the IEEE generative AI authorship policy and foundational peer review principles before preparing conference submissions.

FAQ

What is the difference between an ICML position paper and a main track paper?

Main track papers report completed research, algorithmic innovations, theoretical proofs, or novel empirical benchmarks. Position papers argue a normative stance on what the ML community should do, addressing meta-level perspectives, ethical standards, evaluation criteria, or governance frameworks.

Does an ICML position paper require empirical benchmark experiments?

No. Position papers are not required to introduce new algorithms or benchmark tables. However, claims must be supported by sound reasoning, formal literature citations, or relevant observational data to satisfy standard ICML scholarship requirements.

Can an ICML position paper include an appendix?

Yes, supplementary materials and appendices are permitted for non-essential background details. However, core arguments and the mandatory Alternative Views section must reside entirely within the 8-page main body.

How are ICML position papers reviewed?

Position papers undergo double-blind peer review based on seven criteria: position clarity, evidential support, significance, discussion potential, alternative views, communication quality, and literature context. Reviewers evaluate whether the argument warrants community discussion, independent of personal agreement.

Sources

International Conference on Machine Learning (ICML). (2025). “ICML 2025 Call For Position Papers.” https://icml.cc/Conferences/2025/CallForPositionPapers (read 2026-11-07).
Conference on Neural Information Processing Systems (NeurIPS). (2025). “NeurIPS 2025 Call for Position Papers.” https://neurips.cc/Conferences/2025/CallForPositionPapers (read 2026-11-07).