Roadmap · June 2026 → June 2027
2027 USAAIO Timeline
A ~30-week plan built around a beginner starting in June 2026, studying 10+ hrs/week, aiming to qualify for Round 2. 为期约30周的备赛计划，目标晋级第二轮。
Harper is eligible. USAAIO is open to K–12 students in the U.S. and Canada, and Round 1 is open to everyone. Contestants must be under 20 on the first day of the international competition and not a full-time university student.
Born Dec 28, 2014, Harper will be 12 (turning 13) during the 2027 cycle — far under the age cap. Having finished Algebra 2 is a strong base; the main gap to close is programming (starting from scratch) and the AI-specific math (linear algebra, probability, calculus, optimization). Harper 完全符合参赛资格；主要需要补的是编程和 AI 数学基础。
Jun – Jul 2026 · Module 0 + Math start
Python from zero → data tooling.
Python syntax, control flow, functions, then NumPy, pandas, matplotlib, seaborn. In parallel, begin the math track on Math Academy (pre-calc → linear algebra foundations). Goal: Harper can write clean Python and manipulate arrays/dataframes comfortably.
Aug 2026 · Module 1 + 2
Math foundations for AI + classical ML begins.
Linear algebra (vectors, matrices, eigen-stuff), probability & stats (Bayes, distributions), derivatives & gradient descent. Start supervised learning: linear & logistic regression — both the math derivation and coding from scratch with NumPy, then scikit-learn.
Sep 2026 · Module 2 (deep)
Classical machine learning core.
SVM, decision trees, kNN, ensemble methods, bias-variance tradeoff, cross-validation, loss functions. Emphasis on deriving estimators by hand AND implementing them. Weekly mini-projects on real datasets.
Oct 2026 · Module 3 + 4
Unsupervised ML + deep learning foundations.
k-means and PCA from scratch (eigenvalue problem). Then the multi-layer perceptron: affine layers, batch norm, dropout, and forward/backprop computed by hand. Introduce PyTorch.
Nov 2026 · Module 4 + 5
PyTorch fluency + convolutional neural networks.
Build and train MLPs in PyTorch; then CNNs for image tasks (the Round-1 deep-learning topic). Lots of hands-on Colab notebooks running on CPU.
Dec 2026 · Review + mock exams
Consolidate & simulate.
Work all available past problems. Full 3-hour timed mocks in Colab. Patch weak spots. Confirm registration, proctor site, and Edvistas account are set up. (Harper turns 12 on Dec 28 🎂)
Late Jan 2027 · 🎯 Round 1
Competition day.
3-hour proctored online exam, Google Colab, CPU only. Final week = light review, rest, logistics, and a calm warm-up notebook.
Feb 2027 · Module 6
(If qualified) Transformers & attention.
Attention mechanism, transformer architecture, vision transformers, GNNs. This is flagged as needing "very solid and deep" understanding.
Mar 2027 · Module 6 + 7
NLP + generative AI.
Tokenization, embeddings, pre-training, fine-tuning. Autoencoders, VAEs, GANs, diffusion / stable diffusion. Start GPU work on Colab (L4).
Mar–Apr 2027 · 🎯 Round 2 (MIT)
Everything on the syllabus, GPUs allowed.
Timed mocks on full-syllabus problems. Target: strong showing → camp invitation.
Jun 2027 · Camp (Harvard)
Stretch goal.
If Round 2 qualifies — selection camp, then potential Team USA for IOAI / IAIO.