FAIR Universe: NeurIPS 2026 Weak Lensing Uncertainty Challenge

Winners

Phase 2 of the NeurIPS 2025 Weak Lensing ML Uncertainty Challenge has come to an end. After the careful final evaluation on the holdout dataset, we are pleased to announce the following winners:

Medal Prize Winner Holdout Score Recognition
Gold 1st Place junlong 0.472572 The best overall performance on unseen OoDs
Silver Tie 2nd Place sk7524, apurani 0.446880, 0.440640 The second best overall performance on unseen OoDs
Medal Jury Prize thylinao 0.402456 Recognized for the outstanding generalizability to unseen OoDs
Medal Jury Prize dkn20 0.338694 Recognized for the novelty of their OoD detection approach

The holdout dataset contains 10,000 samples, where the number of InD and OoD maps are both 5,000. The complete final evaluation results, including detailed performance comparisons on individual OoD categories and the score uncertainties from the composition of holdout samples, are available on Phase-2 Final Results Summary.

We will award cash prizes to all the prize winners:

  • junlong: Junlong Shen (University of Albterta, Canada) will win $2,000.
  • sk7524: Sachin Konan (Princeton University, USA) will win $1,000.
  • apurani: Team DONER — Cheng-Tse Huang, Ian Peng, Agni Purani (Rutgers University, USA) will win $1,000.
  • thylinao: Maksim Silchenko (National University of Singapore, Singapore) will win $500.
  • dkn20: Kangning Diao (University of California, Berkeley, USA) will win $500.

All winners will be invited to present at our NeurIPS 2026 Competition Workshop in Paris:

  • Time: December 12th (Saturday) 3:30 PM -- 5:45 PM (CET)
  • Location: Room 362-363, Palais des Congrès
  • Agenda & talks: To be announced soon

We will prepare a paper documenting the top solutions with the winning teams after the workshop. The dataset will also be released permanently to serve as a permanent benchmark.

Congratulations to all the winners! Finally, we would like to thank all participants once again for making this challenge a success. Your contributions are crucial for advancing the exploration of robust and reliable machine learning methods for weak lensing cosmology.