Checklist EUROAI (CEOAI) 2026 Practice Round 1 · Task 2
Project KRAKEN
From image slices, two-channel wave signals and symbolic 'glyph' sequences, predict warping-spline coefficients, a topological class with an open-set label, and a stability index.
The task
A deliberately obscure science-fiction framing describes entities observed through three noisy modalities: 3-channel 128×128 'holographic slices', 1,024-step two-polarisation 'echo' signals with zeroed gaps and time warping, and 'glyph' token sequences.
Subtask 1 (30 pts) predicts the 10 spline coefficients of the time-warping function; Subtask 2 (35 pts) predicts a class in 0–49 or −1 for unseen ('abyssal') forms; Subtask 3 (35 pts) predicts a stability index in [0, 1].
Abridged by SOTA from the official materials. The official statement has the exact rules, and it wins wherever this summary differs.
At a glance
- You get
train/test_slices.npy(N, 3, 128, 128),train/test_echoes.npy(N, 1024, 2),train/test_glyphs.csv,train_targets.csv.- You submit
- One CSV with subtaskID, datapointID, answer: 10 semicolon-separated floats (subtask 1), an integer (subtask 2), a float in [0, 1] (subtask 3).
- Scoring
- Subtask 1: MSE, 30 pts at ≤ 0.00085, 0 at ≥ 0.020. Subtask 2: macro F1 with false positives of class −1 counted double in its precision, 35 pts at ≥ 0.85, 0 at ≤ 0.55. Subtask 3: RMSE, 35 pts at ≤ 0.035, 0 at ≥ 0.100. Linear interpolation in between.
- Rules
- Submission limit 10, two final submissions (platform settings).
- Format
- Online Practice Round 1, 15 March 2026, 13:00–19:00 UTC (6 hours), open to everyone on Nitro AI Judge; tasks proposed by volunteers and reviewed by the International Scientific Committee.