United States · Grades 9 to 12 · AI olympiads and AI projects
Machine learning competitions for high school students
Machine learning competitions for teenagers come in two very different shapes. One is the olympiad: a proctored exam where a student derives a gradient by hand and trains a small model in Google Colab under a clock. The other is the project challenge: a team spends weeks applying a model to a real problem and explains it in writing and on video. A US high schooler can do both in 2026-27. This guide describes each route from its organiser’s own pages, checked on 17 October 2026, and then builds a tiny classifier from scratch to show the level of understanding the olympiad expects.
Live teaching since 2020 · 10,000+ students · models built and explained from first principles
In short
The main machine learning competitions for US high schoolers in 2026-27 are the USA-NA-AIO (an individual, proctored AI olympiad; Round 1 on February 19, 2027, the route to IOAI 2027 in Singapore) and the Presidential SI Challenge (team AI projects; register by December 18, 2026). AI apps can also enter the Congressional App Challenge or a science fair. Modern Age Coders teaches the machine learning behind them live online.
Learn it with a teacher
Three courses that cover the machine learning these contests test
From the Python and maths foundations up to models a student can train and defend.
Five places a US high schooler can compete with machine learning
They differ in one big way: whether the student is tested on understanding, or judged on a project. Every row comes from that organiser’s own pages.
Machine learning and AI competitions open to US high schoolers, 2026-27 (official pages checked 17 October 2026)
Competition
Type
Solo or team
Who can enter
Next milestone
USA-NA-AIO (run by USAAIO)
Proctored olympiad: theory and coding
Individual
K-12 students in the US and Canada who meet the age and status rules
Round 1, Friday, February 19, 2027
IOAI 2027
International olympiad week
National team
High school students chosen by their country; for the US, through the USA-NA-AIO
July 4 to 10, 2027, Singapore
Presidential SI Challenge
Project, written report and video
One to four students and a Team Leader
High schoolers in grades 9 to 12
Registration deadline December 18, 2026
Congressional App Challenge
An app, which may use AI, with a demo video
One student or a team of four at most
Students in middle or high school whose district takes part
The 2026 deadline is October 26, 2026
Science fairs leading to Regeneron ISEF
Research or engineering project, judged in person
Set by each fair
Through a Society-affiliated fair
Set by each local or state fair
Olympiad or project?
The olympiad suits a student who likes exact answers: deriving an estimator, computing a backpropagation step by hand, or writing a training loop that runs on a CPU within the time. There is one score and it is individual.
Project contests suit a student who likes a messy real problem, can work in a team and can explain a model to a judge who is not a specialist. Accuracy still matters, but so do the data, the safety thinking and the presentation.
Why both are worth doing
Olympiad preparation forces the theory. Project work forces the judgment about when a model is good enough and when the data is the real problem.
The calendars barely clash: SI Challenge projects are due in December, and the olympiad’s first round is in February.
The USA-North America AI Olympiad is run by USAAIO, which also organises Team USA for the international AI olympiads. Its 2027 rules were posted on September 9, 2026.
USA-NA-AIO 2027 key facts, from the USAAIO site (checked 17 October 2026)
Item
As USAAIO publishes it
Eligibility
Age: under 20 years old by August 30 of the competition year. Status: a US or Canadian citizen or permanent resident, or a full-time middle or high school student studying in the US or Canada. Not a full-time university student
Format
"It is an individual competition." Every round is taken onsite with a proctor
Registration
June 1, 2026 to January 31, 2027
Round 1
Friday, February 19, 2027, from 12:00 p.m. to 3:00 p.m. U.S. Eastern Time, at schools or approved test sites
Part 1
Non-coding, multiple choice or fill in the blank, 60 minutes, with the contest app locked down
Part 2
Coding, 75 minutes, written and run in Google Colab
Hardware
"All code must run on CPUs. In Round 1, GPUs are neither required nor allowed."
Monitoring
Required monitoring software, a single monitor, and an OBS Studio recording of the full screen and the student’s face
After Round 1
Round 2 in March or April 2027 for qualifiers, some problems may need GPUs; a camp in June 2027
Where it leads
USAAIO says the USA-NA-AIO is the only pathway for students to qualify for Team USA at the international AI olympiads. IOAI 2027, the fourth International Olympiad in AI, is set for July 4 to 10, 2027 in Singapore.
USAAIO also sends a team to the International AI Olympiad (IAIO); its 2027 edition is listed for Vietnam from February 25 to March 2, 2027, with that team already chosen.
Practical points for families
Students register themselves, then find a proctor; the proctor’s details are only collected during the contest. A school teacher, or an approved site such as a library or testing center, can host.
Because the coding part is in Colab with recording software running, a student should rehearse the whole setup on the same laptop at least once before February.
USAAIO publishes a syllabus for the whole competition and says which parts Round 1 covers. It is a useful map of what high school machine learning looks like at contest level, whether or not a student enters.
USA-NA-AIO syllabus areas and the Round 1 subset (USAAIO syllabus and 2027 rules, checked 17 October 2026)
Area
Examples USAAIO lists
In Round 1?
Markdown in Google Colab
Text cells, code snippets and typeset maths
Yes
Mathematical foundations
Linear algebra, Bayes’ rule and Hoeffding’s inequality, multivariable derivatives, gradient descent and duality
Deep learning answers "must be programmed with PyTorch rather than TensorFlow", in line with IOAI. A student who learned TensorFlow first should switch early.
USAAIO also expects theory and code together. Its own example is that knowing the theory without being able to build a PCA class from scratch in NumPy is not enough, and the reverse is not enough either.
How we read it for a grade 10 student
The maths is the long pole. Most of the Round 1 list is reachable for a strong grade 10 or 11 student, but eigenvalues and multivariable derivatives usually have to be taught ahead of the school curriculum.
In USAAIO’s words, the AI Olympiad "is neither a pure math Olympiad nor a pure coding Olympiad". Our plan below splits each week between the two for that reason.
Project contests: the Presidential SI Challenge, apps and science fairs
Here the model is a means, not the test. Judges want to know what problem it solves, whether it works, and whether the team understands its limits.
The Presidential SI Challenge for high school teams, 2026-27 (eligibility, tiers, requirements and timeline pages, checked 17 October 2026)
Point
What the official pages set out
Category
Grades 9 to 12, entered with a parent’s consent
Team
1 to 4 high school students and one Team Leader, who must be 21 or older
Tracks
Standard, or Trailblazer for projects with nationwide impact that are also novel
Hand-in
One PDF: a title page and up to 10 further pages in 12-point type or larger, plus a video of up to four minutes
Original work
All content created specifically for the Challenge after August 1, 2026
Judging
At least three judges per eligible project, using rubrics; State Champions by state, category and track, with two each for California, Florida, New York and Texas
Dates
Team Leaders must register teams by December 18, 2026, and the project upload deadline is December 23, 2026 (both 5:00 p.m. Eastern). State Champions are named between February 11 and 15, 2027; regional judging runs from March 4 to April 8, 2027; the national round is in Washington, DC from June 6 to 9, 2027
The question that tests ML understanding
Among the questions the report must answer is how the team tested or verified the accuracy of any SI-generated content, outputs or predictions, and how accurate they were. That is a machine learning evaluation question in plain words.
A strong answer names the test data, explains why it was kept separate, gives a number, and says where the model fails. The kNN example below shows the smallest version of that habit.
Two other doors for an ML project
The Congressional App Challenge accepts apps built with AI, as long as every use of AI is declared in the entry. Our guide to that challenge covers the rest of its rules, and our Presidential AI Challenge guide goes deeper on the SI Challenge.
Science fairs are the research route. Regeneron ISEF 2026 used 22 categories, including Robotics and Intelligent Machines (ROBO), Software Design (SFTD) and Computational Biology and Bioinformatics (CBIO); each affiliated fair sets its own categories.
The program now called the Presidential SI Challenge was launched as the Presidential AI Challenge. Its official site explains that the federal term changed from "Artificial Intelligence" to "Super Intelligence" under a September 29, 2026 executive order, so search results may show either name. Sources: requirements, competition tiers, important dates, ISEF categories.
A model from scratch
Build a k-nearest-neighbours classifier in NumPy
Olympiad problems expect students to build simple models without a library doing the thinking. This example is ours: a complete kNN classifier in a few lines, on six made-up points, runnable on any laptop CPU.
kNN from scratch with NumPy (ours)Python
import numpy as np
def knn_predict(X_train, y_train, X_new, k=3):
# distance from every new point to every training point
d = np.linalg.norm(X_new[:, None, :] - X_train[None, :, :], axis=2)
nearest = np.argsort(d, axis=1)[:, :k] # indices of the k closest
votes = y_train[nearest] # their labels
return (votes.mean(axis=1) > 0.5).astype(int)
X_train = np.array([[1.0, 1.0], [1.5, 2.0], [2.0, 1.2],
[6.0, 5.5], [7.0, 6.0], [6.5, 7.0]])
y_train = np.array([0, 0, 0, 1, 1, 1])
X_new = np.array([[1.8, 1.5], [6.2, 6.1], [4.0, 3.8]])
print(knn_predict(X_train, y_train, X_new, k=3)) # [0 1 0]
print(knn_predict(X_train, y_train, X_new, k=1)) # [0 1 1]
Run with Python 3 and NumPy. The two outputs differ only in the third point, which sits between the two clusters.
Distances from the in-between point (4.0, 3.8) to each training point (ours, rounded)
Training point
Label
Distance
Rank
(6.0, 5.5)
1
2.62
1
(1.5, 2.0)
0
3.08
2
(2.0, 1.2)
0
3.28
3
(7.0, 6.0)
1
3.72
4
(6.5, 7.0)
1
4.06
5
(1.0, 1.0)
0
4.10
6
What the example teaches
With k = 1, the single closest point has label 1, so the prediction is 1. With k = 3, the three closest are labels 1, 0 and 0, so the majority vote is 0. Same data, same code, different answer, because of a setting the student chose.
That is the core idea behind cross-validation, which is on the Round 1 syllabus: choose k by testing on data the model has not seen, not by what looks right.
Three extensions to try
Scale the second feature by 10 and run it again. Why do the predictions change, and what does that say about normalising features?
Replace the vote with a distance-weighted vote, then write the same classifier as a small PyTorch function instead of NumPy.
Hold out two of the six points, choose k on the other four, and report accuracy on the held-out two.
Eighteen weeks
An 18-week plan to USA-NA-AIO Round 1
Our plan for a student starting in mid-October 2026 with solid Python and algebra. About four hours a week, one of them a live class. February 19, 2027 is roughly 18 weeks away from the date this page was checked.
From October 2026 to Round 1 on February 19, 2027 (our plan, built on the published syllabus)
Weeks
Maths
Code
Done when
1 to 3
Vectors, matrices, dot products
NumPy arrays and broadcasting
The kNN example runs and the student can explain each line
4 to 6
Derivatives and gradient descent
Linear and logistic regression from scratch
A loss curve that goes down, plotted in matplotlib
7 to 9
Probability, Bayes’ rule
scikit-learn models, cross-validation
k chosen by validation, not by guessing
10 to 12
Eigenvalues, PCA
A PCA class in NumPy; k-means
PCA output matches scikit-learn on the same data
13 to 15
Backpropagation by hand
PyTorch MLPs, dropout, batch norm
One backward pass computed on paper and checked in code
16 to 17
Review
Basic CNNs in PyTorch on CPU
A small image model trained inside the time
18
Rest and rehearsal
Full mock in Colab with recording software on
Setup tested on the contest laptop
Grade 9
Start the plan for next year instead; use this year for Python, NumPy and algebra.
Grades 10 and 11
The plan above, plus an SI Challenge project in December if time allows.
Grade 12
Check the under-20 and not-at-university rules; a project contest may fit better.
Where teens start
Four levels between using AI and competing in it
A teen new to ML competitions is usually at level 1 or 2.
From AI user to ML competitor (ours)
Level
In practice
1. User
Uses AI apps; has not trained a model
2. Library user
Can call scikit-learn fit and predict on a dataset
3. Builder
Can write a simple model from scratch and explain the maths
4. Evaluator
Can show, with held-out data, how good a model is and where it fails
For parents
Level 2 to level 3 needs maths that many schools teach later. That is normal, and it is the part we spend most time on.
Ask your teen how they know their model works. A clear answer is a better sign than any accuracy number.
For students
Write one model from scratch before using a library for it. You will understand the library far better afterwards.
Keep a notebook of every experiment, including the ones that failed. Project judges ask about those.
The catalogue
Nine courses for teens heading into ML competitions
Live online, one steady teacher per class, in a group or one to one.
A weather forecasting site with live conditions for any location.
by Krish
Fees
Fees
The competitions set their own entry terms; we charge only for teaching. Lessons are paid monthly in US dollars and can stop at any month end, which suits a block of preparation before a round.
"The one step solution for my son. Modern Age Coders make learning coding so simple that kids love it. The teachers explain complex concepts clearly with practical exercises and interactive content."
Ria Mukherjee
Parent
★★★★★
"Modern Age Coders has been a game-changer for me. I struggled to grasp IT concepts and coding before joining, but their classes transformed everything. I can now confidently write complex programs with ease."
Samriddha Mondal
Student
★★★★★
"One of the most wonderful education centres out there. Education is not limited to school syllabus but focuses on skill development."
Vansh Agarwal
Student
★★★★★
"My child Dhairya is really enjoying the Modern Age Coders classes. This is his first online class and he eagerly looks forward to it. I can already see his improvement, and the teachers are very cooperative."
Sonam Oswal
Parent of Dhairya
★★★★★
"Modern Age Coders have wonderful teachers who teach in a clear, easy and practical way. The teacher boosts students' confidence and inspires them to learn without hesitation."
Sonu Goyal
Parent
★★★★★
"I highly recommend this computer coding class! The teachers are incredibly knowledgeable and passionate about coding."
Ritu Kedia
Parent
Questions from families
ML competitions for teens: what families ask
Which ML contests are open to US high schoolers?
In 2026-27 the main ones are the USA-NA-AIO (the AI olympiad that selects Team USA for IOAI) and the Presidential SI Challenge for team projects. AI apps can also go to the Congressional App Challenge, and research projects to science fairs on the way to Regeneron ISEF.
When is the 2027 USA-NA-AIO?
USAAIO has scheduled Round 1 for Friday, February 19, 2027, from 12:00 p.m. to 3:00 p.m. U.S. Eastern Time. Registration runs until January 31, 2027, and Round 2 is in March or April 2027.
Who can take the USA-NA-AIO?
A US or Canadian citizen or permanent resident, or a full-time middle or high school student studying in the US or Canada, who is under 20 by August 30 of the competition year and is not a full-time university student.
Do I need a GPU for the AI olympiad?
Not for Round 1. USAAIO says all Round 1 code must run on CPUs and GPUs are neither required nor allowed. Some Round 2 problems may require GPUs.
Should I learn PyTorch or TensorFlow?
For the USA-NA-AIO, PyTorch. Its syllabus says deep learning problems must be programmed with PyTorch rather than TensorFlow, in line with IOAI.
Where is IOAI 2027?
The IOAI site lists the fourth International Olympiad in AI for July 4 to 10, 2027 in Singapore. US students reach it through the USA-NA-AIO.
Is the Presidential AI Challenge the same as the Presidential SI Challenge?
Yes. The official site says the name changed after a September 29, 2026 executive order. High school teams of one to four register by December 18, 2026 and submit by December 23, 2026.
What times do your classes run?
Weekdays or weekends, flexible: choose your batch, or for a very personalised plan choose 1-on-1 at your convenience, weekend mornings or weekdays.
What about Kaggle competitions?
Kaggle pages could not be read as plain text when we checked, so we do not describe their rules here. Each Kaggle competition has its own rules page; read the eligibility section with a parent before entering.
Are you part of USAAIO or the SI Challenge?
No. We have no link with any organiser. We teach the machine learning, maths and Python these competitions draw on; registration is always directly with the organiser.
Ten AI projects for US middle schoolers, two in real Python, and how to take one to a science fair.
Start here
Book a free first lesson
Share what your teen has built or studied in AI so far. The teacher works through one small model with them and maps the gaps against the olympiad syllabus or a project idea.
Our teachers teach from India over live video, so classes are online only. Families in the US usually reach us fastest on WhatsApp.
Want a class sooner? Book a priority demo
A priority demo is a full class of about 45 to 60 minutes. After the class the mentor sends a written skill report and a personal learning roadmap on WhatsApp. If you enrol, the demo fee is adjusted against your first month's fee. Share your demo Order ID with us on WhatsApp and we take it off.
