A reinforcement course where F1·F2 graduates adapt to the exam format and math depth of USAAIO Round 1. It focuses on two areas: (a) Writing math answers in Google Colab markdown, (b) by-hand math derivation(computing backpropagation directly, deriving estimators, proving kernel validity). USAAIO eliminates both the "student who only knows theory" and the "student who only knows code," so we train the two abilities separately.
Published: May 16, 2026 | Last updated: May 16, 2026 · Based on official USAAIO information
Track
USAAIO-only
Round 1 math·format reinforcement
Target
F1·F2 graduates
Holds USAAIO eligibility (U0)
Recommended Hours
About 10 hours
Adjusted by class format · varies 6-14 hours
USAAIO mapping
Round 1 reinforcement
Areas not covered by F1·F2
After finishing U1, the student can write math derivations accurately in the USAAIO Round 1 answer format (Colab markdown)and can derive backpropagation·estimators·kernel validity by hand. The goal is not library calls, but the level of developing theory on paper and verifying the result with from-scratch code.
U1 does not re-teach ML itself. On top of the classical ML·deep learning learned in F1·F2, within the USAAIO Round 1 scope it reinforces only the areas not covered by F1·F2 . There are two core operating principles: standardizing Colab markdown answers(to prevent time loss on exam day) and training by-hand calculation vs. code separately(checking the theory and implementation abilities individually).
F1·F2 Completion (or concurrent enrollment) is Required, and you U0 eligibility assessmentto confirm USAAIO eligibility. For the full track structure, see USAAIO Prep Curriculum Hub.
Below is the standard plan on a 1:1 basis. Depending on the student's prior knowledge and absorption speed, some weeks are accelerated, condensed, or covered in greater depth. Core tools: Google Colab, LaTeX, NumPy, PyTorch.
| Week | Topic | Key Deliverable |
|---|---|---|
| 1 | Intro to Colab markdown: LaTeX equations, code cells, text cells | Colab "math answer template" notebook |
| 2 | Advanced linear algebra: affine transformations, matrix decomposition, eigenvalue by-hand calculation | By-hand calculation + 5 Colab answers |
| 3 | Probability·statistics: applying Bayes' theorem, Hoeffding's inequality | 5 by-hand calculations |
| 4 | Multivariable differentiation: Jacobian, Hessian, chain rule | 5 by-hand calculations |
| 5 | Convex optimization: gradient descent convergence proof, duality | By-hand calculation + short PyTorch comparison |
| 6 | Linear regression estimator derivation (closed-form, OLS) | NumPy from-scratch implementation |
| 7 | Logistic regression + cross-entropy differentiation | NumPy from-scratch implementation |
| 8 | Backpropagation by hand: forward/backward of a small network by hand | By-hand answer + PyTorch verification |
| 9 | PCA eigenvalue problem derivation: kernel PCA, kernel validity proof | NumPy from-scratch PCA |
| 10 | Integration: solving math-derivation problems from past USAAIO exams | Colab answers for 5 problems |
※ Weeks are content units; actual time varies by student. Recommended about 10 hours, ranging 6-14 hours.
Each week you write a Colab markdown answer notebook matching the official USAAIO format. By automating mixed LaTeX-equation + code-cell answers into a standard template, you eliminate format-related losses on exam day.
You accumulate an answer set developing backpropagation·estimator derivation·kernel validity proofs by hand. You cross-verify by-hand results with from-scratch code to check the theory and implementation abilities separately.
USAAIO evaluates the competition submission, not CIT's course portfolio. Student-authored derivations and from-scratch implementation notebooks can document the student's actual reasoning, code, results, and limitations. No university publishes a fixed admissions value for these materials.
GitHub
an organized repo usaaio-math-foundations
Colab
10 USAAIO-format answer notebooks
By-hand calculation set
Collection of student-authored derivation answers
Before reusing any work in another project or application, check duplicate-submission, assessed-work, and outside-support rules and disclose prior use where required.
U1 is the first learning course of the USAAIO-only track (after the U0 eligibility diagnostic). For the full track structure, see USAAIO Prep Curriculum Hub.
Current Course
U1. Math + Colab Reinforcement
Round 1 math and format
Yes. U1 assumes you have completed F1 (classical ML)·F2 (neural networks & deep learning) or are taking them concurrently. That's because U1 doesn't re-teach ML itself: it only reinforces the parts of the USAAIO Round 1 scope that F1·F2 don't cover (by-hand math derivations + Colab markdown answer format). You must also have passed U0, having completed the USAAIO eligibility diagnostic.
U1 maps to the parts of the official USAAIO Round 1 scope not covered by F1·F2. It covers the Colab markdown answer format; advanced linear algebra·probability and statistics·multivariable differentiation·convex optimization; derivation of linear/logistic regression estimators; backpropagation by hand; and kernel PCA validity proofs. It focuses intensively on 'the ability to derive theory by hand' and 'the ability to implement from scratch.'
No. The ~10 hours is a recommended figure on a 1:1 basis and varies ±30-50% depending on the student's prior knowledge and pace of understanding (roughly 6-14 hours). A strong math background may need less than the recommended time, while a first encounter with by-hand derivations may need more. We calculate an individual time plan per student in the first diagnostic session.
After U1 comes U2 (Round 1 Mock Bootcamp). The Colab markdown answer templates and by-hand derivation skills you standardized in U1 are automated in U2's time-pressured mock exam environment. After passing Round 1, it continues to U3 (Round 2 Bootcamp), which also requires the A1·A2 advanced courses.
USAAIO itself evaluates the competition submission, not CIT's course portfolio. Student-authored derivation notes and implementation notebooks can be retained as truthful learning records. Any reuse must follow the competition, school, and application platform's rules; no university publishes a fixed admissions value for them.
First we confirm USAAIO eligibility with the U0 diagnostic, then we individually design a U1 time plan in a diagnostic session tailored to your child's math background.