Checklist AICC 2025 Round 0 · Task 3
Latent Model Classification
Decide, for each input–output pair, which of two partially known neural networks produced the observed logits, without any source labels.
The task
Each sample's 5-dimensional logits Y were produced by one of two hidden models, A or B. Both map 100-dimensional inputs to 10-dimensional embeddings with different learned transformations and then apply a hidden final linear head.
Contestants receive the two models up to, but not including, the final layer, together with the inputs X and the observed logits Y. The source of each row is not provided; the problem is intentionally unlabelled.
Contestants predict, for each row, whether model A or model B generated the logits.
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.csv(ID, X_1–X_100, Y_1–Y_5) andmodelA_penultimate.pth,modelB_penultimate.pth(weights up to the 10-dimensional penultimate layer).- You submit
submission.csvwith columns ID and Source ('A' or 'B').- Scoring
- Accuracy of the predicted source labels.
- Rules
- 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 0, online on Kaggle, 4 Oct 2025 18:00 UTC – 5 Oct 2025 18:00 UTC.