Preparation guide
How to prepare for International Olympiad in AI
The newest international science olympiad, with a scientific round and a practical machine learning round.
Elite · High school students living in the country they represent
Where the points actually are
IOAI preparation is genuine machine learning practice rather than exam cramming. The strongest preparation is building and training real models, reading the standard textbooks, and working competitive machine learning problems, because the practical round is closer to a Kaggle-style task than to a written exam.
Topic breakdown
- 01Mathematics
- Linear algebra, probability, calculus for backpropagation, and optimisation. Both rounds assume all of it.
- 02Core machine learning
- Regression, classification, regularisation, cross-validation, bias and variance. The unglamorous foundation that decides practical round scores.
- 03Deep learning
- Convolutional and transformer architectures, training dynamics, fine-tuning, evaluation. Implement these yourself rather than only using libraries.
- 04Practical tooling
- Python, PyTorch, data loading, GPU use, and debugging a training run that is not converging.
- 05Reading papers
- The scientific round poses open-ended problems that reward familiarity with current literature.
A study plan that works backwards from the date
- A year out Work through a full machine learning course such as the free fast.ai or Andrew Ng material, and implement everything rather than watching passively.
- Six months out Build three projects end to end from raw data to evaluated model. This is what the practical round tests.
- Three months out Practise on competitive machine learning problems under time pressure, and strengthen the mathematics behind whatever you use as a black box.
- Before the qualifier Work your national organiser's published sample problems. Formats vary considerably by country.
Tips that come from the format itself
- Implement models from scratch at least once. Understanding backpropagation by having written it is worth more than any amount of library familiarity.
- The practical round is timed and compute-limited. Practise making a working baseline fast before trying to optimise.
- Free high-quality courses exist. fast.ai and the Deep Learning Specialisation cover most of what is needed at no cost.
- This is a young olympiad and formats have changed between editions. Read your national organiser's current material rather than older guides.
Practice resources
-
IOAI official site
Format, syllabus, national organiser directory and past problems.
-
fast.ai Practical Deep Learning
Free, project-first deep learning course. The fastest route from beginner to training real models.
-
Dive into Deep Learning
Free interactive textbook with runnable code covering the full modern syllabus.
-
Kaggle Learn
Free short courses on the practical machine learning workflow.
-
USA Artificial Intelligence Olympiad (USAAIO) contest rules
The US and Canadian national pathway to Team USA. Registration for 2027 closes 31 January 2027.
Past papers and official materials
- IOAI past problems
The organisers publish problems from previous editions.
Videos worth your time
- 3Blue1Brown neural networks series
The clearest available visual explanation of how neural networks and backpropagation actually work.
Skills this builds
- Python and PyTorch or similar frameworks
- Linear algebra and probability
- Neural network architectures
- Prompt and model evaluation
- Reading machine learning papers
- Team problem solving