Discord

Checklist AICC 2026 Round 6 · Task 1

Classifier Classifier

Determine which of 10 classes each of 200 small binary image classifiers was trained to recognise, given only their weights and 50 labelled classifiers.

  • Vision
  • Model analysis (classifying trained networks by target class)
Read the official statement

The task

A studio trained 250 judges, each to recognise exactly one of 10 visual patterns and to answer yes or no for any image (25 judges per pattern). The roster was lost; only 50 judges have been re-identified, and the target pattern of the other 200 must be recovered.

Each judge is a small CNN that takes a 1-channel 28 × 28 image with values in [0, 1] and outputs a single logit (positive means the target class was recognised). The architecture is generated from a seed (2–3 convolutional blocks with 8, 16 or 32 channels and kernel size 3 or 5), and the statement gives the code to rebuild each model from its seed and state_dict.

The 50 training classifiers (5 per class) come with their target class; the 200 test classifiers (20 per class, shuffled) do not. The original training data is not provided.

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
models.pt with train_models (50 dicts: seed, state_dict, target_class in 0–9) and test_models (200 dicts: seed, state_dict).
You submit
submission.csv with header id,class and 200 rows; id is the index in test_models and class an integer 0–9.
Scoring
Macro-averaged F1 over the 10 classes. Baseline 0.19, reference solution 1.00.
Rules
  • No external datasets or pretrained models; the original training data is not available.
  • Individual participation (maximum team size 1); at most 15 submissions per day.
  • AICC contest rules (stated on each Kaggle rules page, not enforceable): no use of LLMs for writing code or getting task ideas; no internet use other than official library documentation and the contest platform; no communication with anyone during the contest; clarifications only via the #clarification-requests channel on the AICC Discord server.
Format
AICC Round 6, online on Kaggle, 24 Apr 2026 17:59 UTC – 26 Apr 2026 18:00 UTC.

Notebook

Competition

Solutions

Details

Year
2026, Online (Kaggle)
Round
Round 6 · Task 1
Language
English
License
Varies by task: Classifier Classifier — CC BY-NC-SA 4.0; Massive Problem — 'Subject to Competition Rules' (republishing the task or its data without the authors' permission is forbidden); Nuclei Reconstruction — CC BY-NC 4.0. Solutions repository: MIT., as stated by the source