AI Innovators Challenge

Competition Overview

The AI Innovators Challenge (AIIC) is an individual, two-round competition under the umbrella of the International Artificial Intelligence Innovation Olympiad (IAI²O), consisting of an AIIC Qualifier and an AIIC Final.

Each round features two 90-minute rounds:

• Part 1: AI Theoretical Foundations
• Part 2: AI Practical Challenge

Registeration Period

Registration is from April to July 10, 2025.

Eligibility

The competition is open to individual participants only and has no nationality restrictions. To be eligible, students must be currently enrolled full-time in a secondary school at the Grades 7–12 level, regardless of country.

AIIC Qualifier

Round 1: AI Theoretical Foundations

Participants will solve a series of mathematical, algorithmic, and programming problems related to core AI concepts. This includes mathematical reasoning and equations, as well as intuitive understanding of algorithms and machine learning principles.

Format: 25 multiple choice questions (2 points each) + 10 fill-in-the-blank questions (5 points each)
Programming Language: Python
Total Score: 100 points
Duration: 90 minutes
Focus: Mathematical logic, basic coding, algorithmic thinking, and AI fundamentals

Round 2: AI Practical Challenge

Participants will solve a series of math, algorithmic, and coding problems related to core AI concepts. Questions may involve mathematical reasoning, equations, and intuitive understanding of algorithms and machine learning principles.

Format: One practical problem (computer-based)
Note: Students are required to bring their own laptops for the in-person exam. Limited internet access will be provided during the test. The use of AI tools (e.g., ChatGPT or similar) is strictly prohibited.
Programming Language: Python
Total Score: 100 points
Duration: 90 minutes
Evaluation Criteria: Model performance (e.g., precision, recall, F1 score, ROC & AUC, etc.) and code quality (e.g., clarity, structure).
This round assesses participants’ ability to apply AI knowledge in real-world scenarios and produce functional solutions.

AIIC Final

The competition format will be similar to the AIIC Qualifier structure, but it will include more challenging problems to push participants further. Each part will last for 3 hours. The AIIC Final will be held in person at the Massachusetts Institute of Technology (MIT), Cambridge, US, from September 14-17, 2026.

Syllabus

1. Data Preparation

Data Processing

Python Pandas
Python Numpy
Linear algebra fundamentals
Exploratory data analysis (EDA)

Data Visulization

Python Matpolitlib
Python Seaborn

2. Supervised Learning

Regression Models

Linear regression (Least Squares, Gradient Descent)
Polynomial regression

Classification Models

Decision Trees (entropy, information gain)
Support Vector Machines (SVM) and hyperplanes
Naïve Bayes classification

Regularization & Overfitting

L1 (Lasso) and L2 (Ridge) Regularization
Cross-validation techniques

3. Unsupervised Learning

Clustering

K-means clustering (cost function minimization)
Hierarchical clustering
Gaussian Mixture Models (GMM)

Dimensionality Reduction

PCA and its application in visualization
t-SNE for high-dimensional data representation

4. Reinforcement Learning

Markov Decision Processes (MDPs)

States, actions, rewards
Bellman Equations

Q-Learning

Temporal Difference Learning
Exploration vs. exploitation

5. AI Search

Uninformed Search

Breadth-First Search (BFS), Depth-First Search (DFS)
Uniform Cost Search

Informed Search

A* Algorithm and heuristics
Minimax search (game AI)

6. Logical Reasoning

Propositional & First-Order Logic

CNF, logical connectives
Unification in inference

Resolution Theorem Proving

Satisfiability and SAT solvers

7. Evaluation of ML Models

Classification Metrics

Precision, recall, F1-score
ROC-AUC curves

Bias-Variance Tradeoff

Overfitting detection
K-fold cross-validation

8. Constraint Satisfaction Problems

Backtracking & Constraint Propagation

AC-3 algorithm
Forward checking

9. Kernel Methods

Support Vector Machines (SVM)

Margin maximization
Soft-margin vs. hard-margin SVM

Kernel Trick

Transforming low-dimensional data to high-dimensional space
Radial Basis Function (RBF) kernel

10. Recommender Systems

Collaborative Filtering

User-based and item-based filtering
Similarity metrics (cosine similarity, Pearson correlation)

Content-Based Filtering

TF-IDF and word embeddings
Feature engineering for recommendation models

Hybrid Recommender Systems

Combining collaborative and content-based filtering
Matrix factorization (SVD, ALS algorithms)
Deep learning approaches (Neural Collaborative Filtering, Autoencoders for recommendations)

Resources

Here are some resources that will be useful when preparing for the CAIO contest.

We expect participants who do well in the contest to have some knowledge of the following areas.