About the USA AI Olympiad
USAAIO is an artificial intelligence olympiad for middle and high school students based (or schooled) in the United States and Canada. It runs an annual three-stage contest — Round 1 (online qualifier), Round 2 (in-person, invitation only), and the USAAIO Camp — and selects Team USA for the International Olympiad in Artificial Intelligence (IOAI) and the International AI Olympiad (IAIO).
Format
| Stage | Where | What it tests |
|---|---|---|
| Round 1 — Online qualifier | Remote, Google Colab (CPU-only — GPUs neither needed nor allowed); proctored | Math foundations, Python data wrangling, classical ML, basic deep learning; mix of non-coding/math-reasoning and coding tasks |
| Round 2 — In-person | University venue (MIT for 2026; Harvard in prior years); invitation by Round 1 threshold; Colab with L4 GPU may be used | Modeling problem on a real dataset; theory short-answer section; reproducibility deliverables |
| USAAIO Camp | MIT (June each year); invitation by Round 2 ranking | Team-selection tests; top-ranked campers represent the U.S. at IOAI and IAIO |
Exact dates, venue, and registration window each cycle are posted at usaaio.org — always check there for the current cycle rather than trusting notes like this one.
Eligibility
All three must hold (per usaaio.org contest rules):
- U.S. or Canadian citizen, permanent resident, or a full-time middle/high school student in the U.S. or Canada (U.S./Canadian citizens at schools outside North America also stay eligible).
- Under 20 years old on the first day of the international competition that year.
- Not a full-time university student during the competition year.
No prior contest experience required. Self-study is the typical path; many qualifiers are self-taught from courses + practice.
Allowed tools
- Submission environment: all answers go through Google Colab notebooks (code cells + markdown cells for math). Round 1 is CPU-only; Round 2 may use L4 GPU runtimes.
- Proctoring is mandatory. Un-proctored scores are not certified and cannot advance.
- Language: Python (3.x).
- Required libraries: NumPy, pandas, matplotlib, seaborn, scikit-learn, PyTorch.
- Authorized tools / allowed pre-trained models are listed in a per-cycle Google Drive doc linked from usaaio.org/contest-rules. Read it before each round rather than assuming the prior year's list.
Topic coverage
The official syllabus (usaaio.org/syllabus) breaks down into the strata below. Beyond what's shown, the syllabus also explicitly tests UNet, variational autoencoders (VAE), denoising diffusion (DDPM) and Stable Diffusion, graph neural networks (GNN), plus math items like Hoeffding's inequality, Bayes' rule, matrix decompositions, and convex optimization (gradient descent, duality). Track these even though they don't yet each have their own page.
Foundations
Linear algebra (vectors, matrices, eigen-decomposition), probability and statistics, multivariable calculus, convex optimization.
CodePython & the data stack
NumPy arrays, pandas DataFrames, matplotlib / seaborn for plots, scikit-learn for ML APIs.
Classical MLSupervised & unsupervised
Regression, classification, ensemble methods, cross-validation, clustering, dimensionality reduction.
Deep LearningNeural networks
Multi-layer perceptrons, standard layers, forward / backpropagation, training loops, regularization, optimizers.
Modern AITransformers & applications
Attention mechanism, transformer blocks, NLP (tokenization, embeddings, pre-train / fine-tune), vision (CNNs, detection, autoencoders, GANs, diffusion).
Scoring & promotion
- Round 1 awards (2026 thresholds out of 300): Distinguished Honor Roll (≥201), High Honor Roll (≥146), Honor Roll (≥98). The Honor Roll cutoff (98) also serves as the Round 2 invitation threshold — advancement is threshold-based, not a fixed slot count. (2025 reference: 393 Round 1 participants, ~19.3% advanced.)
- Round 2 awards: USAAIO Campers, Gold Medal, Silver Medal, Bronze Medal, Honorable Mention.
- USAAIO Camp (June, at MIT): Campers take a series of team-selection tests. Top-ranked campers — not Round 2 winners directly — represent the U.S. at IOAI and IAIO.
Why this olympiad is worth your time
- End-to-end ML practice. You learn to take a dataset from messy CSV all the way through a deployable model — exactly the skill the rest of your AI career will demand.
- Forces you to actually understand gradients. Treating PyTorch as a magic black box doesn't survive the final-round theory section.
- Direct international path. IOAI exists, the field is young, and getting onto Team USA at Grade 11–12 is a real possibility for a Grade 9 student who starts now.
- Transferable. Whether or not you ever compete again, the syllabus is a clean curriculum for anyone serious about ML.