Workshop for Autonomous ML Research

The premise

Do conferences exist for science, or for scientists?

The research process comprises the formulation of a hypothesis, the design of an experiment, and the judgment of a result. The implicit assumption that this process is a fundamentally human act is suddenly being challenged by autonomous research.

Given the development of agent harnesses that can pursue long-horizon tasks, we believe it is time for the community to formally recognize and plan for the inevitability of impactful autonomous research.

Our goal is to make autonomous research meaningful in a way that strengthens the ML community without undermining the role of human researchers. The workshop brings this structural change into the open and anchors autonomous research in human judgment and participation.

01

What is the best way to responsibly include autonomous research?

02

How should people evaluate, curate, and take responsibility for AI-generated science?

03

How do we ensure conferences continue to facilitate the development of human researchers?

Call for papers

What belongs here?

We seek ML research substantially carried out by autonomous agents, alongside work on the systems, evaluation methods, and norms that make autonomous research possible.

01

End-to-end autonomous research

A scientific manuscript where an agent carried out research end-to-end or was responsible for the central claims—particularly generating a hypothesis and validating it with experimental or theoretical evidence.

02

Systems for autonomous discovery

Agent harnesses, environments, benchmarks, and infrastructure that enable autonomous ML research loops, replication, or algorithmic discovery.

03

Evaluation, norms, and governance

Research on attribution, disclosure, reproducibility, safety, review, incentives, and the institutional consequences of autonomous research.

A note on scope

The workshop is specifically focused on machine learning research—not autonomous research in unrelated scientific domains—while welcoming relevant lessons from adjacent fields.

Human authorship and AI disclosure

Authorship remains exclusively human. We recognize that the role of author may shift toward curation of autonomously generated research. Human authors nevertheless remain responsible for the paper: they curate the work, verify its claims, disclose agent involvement, and stand behind the final submission.

Disclosure policy pending Final disclosure categories and required supporting artifacts will be posted here.

Review and discussant policy

Since the barrier to producing a paper is lower, the scarce resource becomes human evaluation. Each submission must nominate min(number of authors, 3) qualified reviewers who may be called upon as discussants.

Program chairs will ensure that each accepted submission receives at least three independent reviews. Submissions and reviewers remain anonymous through decision-making; after acceptance, papers are published with their reviews and reviewers. NeurIPS conflict-of-interest policy applies throughout.

Important dates

2026 timeline

July 15Call for papers

August 22Abstract deadline

August 29Final submission

September 24Decisions

September 29Discussant notification

November 13Camera-ready

Exact time zones will be posted with the submission instructions.

Discussant format

Keep human judgment in the foreground.

If autonomous systems can produce candidate papers at scale, the scarce human contribution becomes the ability to evaluate AI-generated claims.

The human author presents a brief summary, then transitions into a dialogue with a human reviewer. The discussant explains why the paper was accepted, what limitations remain, and why its central claim is worthy of attention.

01

Autonomous research

An agent substantially carries out the research loop.

02

Human authorship

Authors curate, verify, disclose, and take responsibility.

03

Blind review

Anonymous evaluation remains separate from the decision.

04

Open discussion

An accepting reviewer joins the author as discussant.

View the proposed one-day schedule

9:00–9:15Opening: What counts as autonomous research?

9:15–9:50Invited talk · Sherry Yang

9:50–10:25Invited talk · Mengdi Wang

10:25–10:45Break

10:45–11:20Invited talk · Nik Dawson

11:20–12:00Spotlight author–discussant sessions

12:00–1:15Lunch and informal discussion

1:15–2:45Spotlight author–discussant sessions

2:45–4:30Poster session

4:30–5:15Town hall

People

Across systems and society.

Mengdi Wang

Princeton University

Professor of Electrical and Computer Engineering. Her work on LabOS explores how multimodal AI agents, extended-reality interfaces, and human–AI collaboration can support physical laboratory science.

Sherry Yang

NYU Courant · Google DeepMind

Assistant Professor and Staff Research Scientist. Her work frames ML engineering as an interactive agent environment and studies how MLE agents improve through experience despite sparse rewards.

Nik Dawson

Burning Glass Institute

Director of Data Science. His research applies ML to labor-market dynamics, including job transitions, skills shortages, and how AI adoption ripples through industries.

Organizers

Arjun Prakash · Aditya Iyer · Hamish Ivison · Jack Liell-Cock · Amy Greenwald · Nora Ayanian

Area chairs

David Tao · Kevin Wang · Stephen Chung · Anna Hakhverdyan · Stephen Crawford · Zarif Aziz

FAQ

Common questions

What counts as autonomous ML research?

ML research substantially produced by an autonomous agent—ideally work where an agent carried out the process end-to-end or was responsible for the central claims, including hypothesis generation and experimental or theoretical validation.

Can humans still be authors?

Yes. Authorship remains exclusively human. Authors curate the work, verify its claims, disclose agent involvement, and remain responsible for the final submission.

Can I submit work about autonomous research systems?

Yes. We welcome papers on systems that produce autonomous research, including environments, harnesses, evaluation methods, replication infrastructure, and governance.

Is the workshop archival?

No. This is a non-archival workshop, allowing authors to continue developing and submitting their work elsewhere, subject to those venues' policies.

Submission

Prepare your work.

OpenReview, formatting, paper length, and final disclosure instructions are coming soon.

Submission details pending

The submission link will appear here when the portal opens.