KOAI and IOAI Syllabus

Syllabus Introduction

The International Olympiad in Artificial Intelligence (IOAI) is a global AI competition for high-school students, aimed at developing both a solid theoretical foundation and hands-on implementation skills. This syllabus outlines the topics that participants should master in order to perform well in the competition. Every year, the IOAI International Scientific Committee (ISC) updates the official syllabus to reflect the latest research trends and educational priorities.

Topic Categories

Topics are divided into three categories according to the nature and level of knowledge participants are expected to have.

Theory (how it works): Participants must understand the core concepts and theoretical background that answer the "why" behind AI algorithms. They study the principles through textbooks, lectures, and so on, prioritizing breadth of understanding over depth.
Practice (what it does, when to use it, how to implement it): Participants must develop the practical ability to implement AI techniques in code. They should be able to use library functions appropriately, apply them to real data, and interpret the outputs. For example: they do not need to dig into the inner workings of the Adam optimizer, but they must be able to judge when and how to use it.
Theory + Practice: Some topics require both theoretical understanding and practical implementation ability.

Subject 1. Foundational Skills and Classical Machine Learning

1-1. Programming Fundamentals

Subtopic	Theory	Practice
Python fundamentals (loops, functions, etc.)	X	O
NumPy, Pandas for data processing	X	O
Matplotlib, Seaborn for visualization	X	O
Scikit-learn for machine learning	X	O
PyTorch fundamentals	X	O
Tensor (multidimensional array) manipulation	X	O
Model training on CPU/GPU	X	O

1-2. Supervised Learning

Subtopic	Theory	Practice
Linear regression	O	O
Logistic regression	O	O
L1, L2 regularization	O	O
K-Nearest Neighbors (K-NN)	O	O
Decision trees	O	O
Ensemble models (Gradient Boosting, Bagging, Random Forest)	X	O
Support Vector Machines (SVM)	O	O

1-3. Unsupervised Learning

Subtopic	Theory	Practice
K-Means clustering	O	O
Principal Component Analysis (PCA)	O	O
t-SNE, UMAP	X	O
DBSCAN, hierarchical clustering, spectral clustering	X	O

1-4. Data Science Fundamentals

Subtopic	Theory	Practice
Model evaluation metrics (accuracy, precision, recall, F1, etc.)	O	O
Underfitting, overfitting	O	X
Hyperparameter tuning	X	O
Cross validation	X	O
Confusion matrix, ROC curve	O	O
Feature engineering*	X	O
Data preprocessing**	X	O

*Feature engineering: transforming high-dimensional, categorical, time-series, or variable-length raw data into a small number of highly informative features. Includes sliding window, pooling, one-hot encoding, statistical-moment-based features (mean, standard deviation), PCA, neural embeddings, and so on.

**Data preprocessing: handling of missing and irregular data (including sequence modeling). Includes basic imputation strategies (mean / median / forward-fill), padding of variable-length sequences, normalization and standardization, train/validation/test split strategies, basic augmentation (flip, crop, noise), text/audio tokenization and vocabulary construction, image patching, and so on.

Subject 2. Neural Networks and Deep Learning

2-1. Neural Networks

Subtopic	Theory	Practice
Perceptron fundamentals	O	O
Gradient descent	O	O
Backpropagation	O	O
Activation functions (ReLU, Sigmoid, Tanh)	O	O
Loss functions (MSE, MAE, Cross Entropy, etc.)	O	O

2-2. Deep Learning

Subtopic	Theory	Practice
Multi-Layer Perceptron (MLP)	O	O
Data embedding (text, image, audio)	O	O
Pooling methods (Max, Average)	O	O
Attention mechanism	O	O
Transformer (theory limited to text and image)	O	O
Autoencoder	X	O
SGD, mini-batch gradient descent	O	O
Momentum-based methods (Adam, AdamW)	X	O
Convergence and learning rate	X	O
Regularization (Dropout, Early Stopping, Weight Decay)	X	O
Weight initialization	X	O
Batch normalization	X	O
Model fine-tuning (full and PEFT)	X	O

Subject 3. Computer Vision

3-1. CV Fundamentals

Subtopic	Theory	Practice
Convolutional layer	O	O
Image classification	X	O
Object detection (YOLO, SSD, DETR)	X	O
Image segmentation (U-Net)	X	O
Pretrained vision encoders (ResNet, etc.)	X	O
Image augmentation	X	O
Image generation with GANs	X	O
Vision self-supervised learning	X	O
Vision-text encoders (CLIP, etc.)	X	O
Diffusion models	X	O

Subject 4. Natural Language Processing and Audio

4-1. For Natural Language Processing

Subtopic	Theory	Practice
Text classification	X	O
Pretrained text encoders (BERT, etc.)	O	O
Language modeling	O	O
Encoder-decoder models (machine translation, vision-language modeling, etc.)	X	O
Pretrained language models (open-source and API-based)	X	O

4-2. Audio Processing

Subtopic	Theory	Practice
Pretrained audio encoders (HuBERT)	X	O
Audio models (Qwen-Audio, Whisper, Voxtral)	X	O

Note: Data can be in any of the forms of text, tables, images, audio, video, or time series, and must be handled with the methods above.

한국정보기술진흥원

KOAI / IOAI Syllabus

IOAI 2026 edition