Checklist MAIO 2025 Malaysia AI Olympiad (national competition) · Task 2
Footprint of a Letter
Invert a small fixed neural network: find the four letters (and their input values) that produced four given output vectors.
The task
A two-layer network SimpleNet (26 -> 64 -> 50, ReLU, weights initialised from N(0, 0.02) with seed 42 and zero biases) takes a letter as a 26-dimensional vector that is zero except for a non-zero value at the letter's position.
Four 50-dimensional output vectors ('footprints') are given. The contestant must find the four letters that produced them, as a single lowercase string, and, for extra credit, the numerical values used to represent the letters.
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
- The notebook defining SimpleNet and the four footprint tensors.
- You submit
- The four letters as one lowercase string, with the notebook showing the method.
- Scoring
- 2 pts for the four correct letters; +2 pts if correct and the notebook shows clear, verifiable reasoning; +3 pts if correct and the method would keep a similar runtime with an alphabet of 10,000 characters; +3 pts if correct and the numerical values representing the letters are found within +/-5%. Partial credit at discretion.
- Rules
- PyTorch 2.x; the random seed must stay fixed at 42.
- Format
- MAIO 2025, the first Malaysia AI Olympiad: online over the weekend of 14-16 March 2025 (48 hours), with objective questions and three handcrafted challenges (maximum total 55 points). This problem was designed to be solved within 2 hours; all LLM assistance, including reasoning models, was allowed. Originally graded with an autograder via API submissions plus hand grading; significantly modified for distribution.