IOAI Resources

Overview

This page provides learning resources aligned with the official International Olympiad in Artificial Intelligence (IOAI) syllabus. NOKI preparation follows this framework to support structured development for national and international AI competitions. As NOKI introduces independent national rounds, the syllabus will be updated and expanded with additional reading resources for participants.

The framework below reflects the IOAI 2026 syllabus. The IOAI International Scientific Committee updates the syllabus each year to track the latest research and educational priorities, so always confirm the current scope against the official source.

Important notice: The topics listed below are intended for extended preparation and are not automatically part of every NOKI competition round. The exact scope for each round is published by NOKI in official announcements.

Topic Classification

IOAI topics are grouped into three categories:

• Theory (how it works): conceptual understanding of principles, methods, and models — prioritizing breadth over depth.
• Practice (what it does, when to use it, and how to implement it): hands-on implementation using programming tools and frameworks.
• Both: topics that require both conceptual knowledge and practical application.

Topics not listed in the official syllabus are considered excluded.

Section 1: Foundational Skills and Classical Machine Learning

Programming Fundamentals

• Python basics: loops, functions, and core syntax (Practice)
• Data handling with NumPy and Pandas (Practice)
• Data visualization with Matplotlib and Seaborn (Practice)
• Machine learning with scikit-learn (Practice)
• PyTorch basics (Practice)
• Tensor (multi-dimensional array) manipulation (Practice)
• Training models on CPU and GPU (Practice)

Supervised Learning

• Linear Regression (Both)
• Logistic Regression (Both)
• L1 and L2 regularization (Both)
• K-Nearest Neighbors (K-NN) (Both)
• Decision Trees (Both)
• Model ensembles: Gradient Boosting, Bagging, Random Forest (Practice)
• Support Vector Machines (SVM) (Both)

Unsupervised Learning

• K-Means Clustering (Both)
• Principal Component Analysis (PCA) (Both)
• t-SNE and UMAP (Practice)
• DBSCAN, hierarchical, and spectral clustering (Practice)

Data Science Fundamentals

• Model evaluation metrics: Accuracy, Precision, Recall, F1-score (Both)
• Underfitting and overfitting (Theory)
• Hyperparameter tuning (Practice)
• Cross-validation (Practice)
• Confusion matrix and ROC curves (Both)
• Feature engineering (Practice)
• Data processing (Practice)

Feature engineering: transforming raw, high-dimensional, categorical, time-series, or ragged data into a compact set of informative features — for example sliding windows, pooling operations, one-hot encoding, statistical-moment features (mean, standard deviation), PCA, and neural-network-based embeddings.

Data processing: handling missing and irregular data, including in sequence settings — basic imputation (mean/median/forward-fill), padding for variable-length sequences, normalization and standardization, train/validation/test splitting, basic augmentation (flipping, cropping, noise), tokenization and vocabulary building for text and audio, and patching for images.

Section 2: Neural Networks and Deep Learning

Neural Networks

• Perceptron basics (Both)
• Gradient descent (Both)
• Backpropagation (Both)
• Activation functions: ReLU, Sigmoid, Tanh (Both)
• Loss functions: MSE, MAE, Cross Entropy (Both)

Deep Learning

• Multi-Layer Perceptrons (MLP) (Both)
• Data embeddings for text, image, and audio (Both)
• Pooling techniques: max and average pooling (Both)
• Attention mechanism (Both)
• Transformers (theory required for text and image) (Both)
• Autoencoders (Practice)
• SGD and mini-batch gradient descent (Both)
• Momentum methods: Adam, AdamW (Practice)
• Convergence and learning rates (Practice)
• Regularization: dropout, early stopping, weight decay (Practice)
• Weight initialization (Practice)
• Batch normalization (Practice)
• Model finetuning: full and parameter-efficient (e.g. LoRA) (Practice)

Section 3: Computer Vision

• Convolutional layers (Both)
• Image classification (Practice)
• Object detection: YOLO, SSD, DETR (Practice)
• Image segmentation: U-Net (Practice)
• Pre-trained vision encoders (e.g. ResNet) (Practice)
• Image augmentation (Practice)
• Generating images with GANs (Practice)
• Self-supervised learning for vision (Practice)
• Vision-text encoders (e.g. CLIP) (Practice)
• Diffusion models (Practice)

Section 4: Natural Language Processing and Audio

Natural Language Processing

• Text classification (Practice)
• Pre-trained text encoders (e.g. BERT) (Both)
• Language modeling (Both)
• Encoder-decoder models (e.g. machine translation, vision-language modeling) (Practice)
• Pre-trained language models (open-source and API-based) (Practice)

Audio Processing

• Pre-trained audio encoders: HuBERT (Practice)
• Audio models: Qwen-Audio, Whisper, Voxtral (Practice)

Data may be text, tabular, image, audio, video, or time-series, and should be processed with the methods above.

Guidance for Participants

• Programming expectations: Prior coding experience is useful but not mandatory. Students with different backgrounds can still participate effectively.
• Typical task formats: Theoretical questions, coding exercises, data analysis, and applied problem-solving scenarios.
• Recommended preparation: Combine syllabus study with practical implementation, regular exercises, and review of official documentation.

For authoritative updates, always refer to the official IOAI 2026 syllabus.