[Competition] NAS Unseen Data Challenge @ AutoML 2026

[Linus Ericsson]

We are happy to invite teams to participate in the NAS Unseen Data Challenge, to be held in conjunction with AutoML 2026!

Website: https://www.nascompetition.com/

The Challenge

Most existing neural architecture search (NAS) methods are tuned for a small number of well-known benchmarks such as ImageNet or CIFAR. In this competition, we raise the bar: can you design a NAS pipeline — including search space, algorithm, and training policy — that generalises to entirely novel, unseen datasets?

Competition Timeline

• Phase 1 (Development): Participants use the provided starter kit to develop and test their method locally. Open now!
• Phase 2 (Validation): Submitted code is run as a “smoke test” on our systems. This phase runs from July 1st to July 31st (making it a great good fit for e.g. summer Master’s projects).
• Phase 3 (Final Evaluation): Validated methods are evaluated on secret datasets to determine the final rankings.

The winners will be announced and celebrated at the AutoML 2026 conference in Ljubljana.

We look forward to your participation and to seeing how well NAS methods can truly generalise beyond familiar benchmarks.

If you have any questions, send us an email: nas-competi...@newcastle.ac.uk

Best regards,

The Organising Team

David Towers, Researcher/PhD Candidate at Newcastle University
A. Stephen McGough, Reader at Newcastle University
Amir Atapour-Abarghouei, Assistant Professor at Durham University
Elliot J. Crowley, Senior Lecturer at the University of Edinburgh
Linus Ericsson, Assistant Professor at the University of Glasgow
Felix Möller, PhD Candidate, Helmholtz-Zentrum Berlin