Syllabus

Updated: 21 September 2026.

1. Topic Categories and Depth Levels

Each topic has a knowledge category (theory, practice, or both) and a depth level (core or extended).

Knowledge Category

Category Meaning
Theory (T) Understand the main concepts, assumptions, mechanisms, and standard equations. Proof-heavy derivations are not expected unless explicitly stated.
Practice (P) Use appropriate libraries and tools, implement or adapt methods, train and debug models, and interpret outputs.
Both (B) Demonstrate both conceptual understanding and practical implementation ability.

Depth Level

Level Meaning
Core (C) May be assessed directly without a task-specific tutorial.
Extended (E) May be assessed when supported by starter material, a provided implementation, or sufficient explanation in the task statement.

2. General Competencies

  • Understand the task, data schema, target, constraints, evaluation metric, and submission format.
  • Inspect data, identify quality issues, establish a simple baseline, and design a suitable validation strategy.
  • Select methods that match the data modality, dataset size, compute budget, and evaluation objective.
  • Implement, train, debug, profile, and compare models using the provided Python environment.
  • Recognize leakage, overfitting, distribution shift, class imbalance, unstable training, and metric misuse.
  • Interpret experiments and make evidence-based improvements under time and compute constraints.
  • Produce deterministic outputs and correctly formatted submissions.

3. Mathematics, Statistics, and Optimization

Topic Expected competency Category Level
Linear algebra Vectors, matrices, tensors, dot products, matrix multiplication, norms, rank, projections, eigenvalues/eigenvectors, and singular value decomposition. B C
Calculus and automatic differentiation Derivatives, partial derivatives, gradients, chain rule, Jacobian/Hessian intuition, computational graphs, and backpropagation. B C
Probability Random variables, common distributions, expectation, variance, covariance, conditional probability, Bayes’ rule, independence, and sampling. B C
Statistics and estimation Likelihood, maximum likelihood, bias-variance trade-off, confidence intervals, bootstrap, and basic hypothesis testing. B C
Information theory Entropy, cross-entropy, KL divergence, mutual information, and their use in learning objectives. B C
Optimization Gradient descent, stochastic and mini-batch optimization, momentum, Adam/AdamW, learning-rate schedules, constrained optimization intuition, and convergence diagnostics. B C
Numerical computation Floating-point behavior, numerical stability, vectorization, memory use, computational complexity, and stable implementations of common operations. B C
Discrete mathematics and graphs Sets, relations, combinatorics, recursion, dynamic programming, graph terminology, paths, connectivity, and adjacency representations. B C
Advanced statistical foundations Bayesian estimation, posterior prediction, multiple testing, and uncertainty quantification at a conceptual and applied level. B E

4. Programming, Data Handling, and Experimentation

Topic Expected competency Category Level
Python programming Control flow, functions, modules, classes, comprehensions, iterators, exceptions, file handling, and standard data structures. P C
NumPy and tensor manipulation Indexing, broadcasting, reshaping, masking, vectorized computation, reductions, and multi-dimensional array operations. P C
Pandas and structured data DataFrames, joins, grouping, aggregation, missing values, categorical variables, time indices, and efficient input/output. P C
Visualization Use Matplotlib and Seaborn to inspect distributions, learning curves, errors, embeddings, predictions, and model behavior. P C
Scikit-learn and boosting libraries Build pipelines and use scikit-learn, XGBoost, LightGBM, or CatBoost appropriately for classical ML tasks. P C
PyTorch fundamentals Tensors, datasets and data loaders, modules, losses, optimizers, device placement, training loops, evaluation mode, and checkpointing. P C
CPU/GPU execution Move data and models correctly, manage device memory, choose batch sizes, use mixed precision when appropriate, and diagnose bottlenecks. P C
Data processing Normalization, standardization, imputation, padding, masking, tokenization, patching, resampling, augmentation, and variable-length data handling. P C
Feature engineering Create compact, informative features from tabular, categorical, image, text, audio, graph, and time-series data. B C
Validation and experimental design Train/validation/test splitting, cross-validation, grouped and temporal splits, ablation studies, and prevention of data leakage. B C
Metrics and error analysis Select, implement, and interpret task-appropriate metrics; inspect confusion matrices, ROC/PR curves, calibration, and subgroup performance. B C
Hyperparameter optimization Manual search, grid/random search, efficient search strategies, early stopping, and fair model comparison. P C
Reproducibility and code quality Random seeds, deterministic behavior, configuration management, logging, assertions, tests, clean notebooks, and reusable functions. P C
Efficient data pipelines Caching, batching, parallel loading, memory mapping, sparse data, and profiling time and memory usage. P E

5. Artificial Intelligence Foundations

Topic Expected competency Category Level
Intelligent agents Agents, environments, observations, actions, goals, utility, rationality, and the perception-action loop. T C
Uninformed search Breadth-first, depth-first, uniform-cost, iterative deepening, completeness, optimality, and complexity. B C
Informed and heuristic search Greedy best-first search, A*, admissibility, consistency, heuristic design, and beam search. B C
Adversarial search Minimax, alpha-beta pruning, evaluation functions, stochastic games, and search-depth trade-offs. B C
Constraint satisfaction Variables, domains, constraints, backtracking, propagation, ordering heuristics, and local search. B C
Planning State-space planning, action models, goals, planning graphs, and the relation between planning and search. B E
Knowledge representation and logic Propositional and first-order logic, inference, rule systems, ontologies, and limitations of symbolic representations. T E
Probabilistic reasoning Bayesian networks, conditional independence, exact and approximate inference, hidden Markov models, and sequential belief updates. B E
Decision theory Expected utility, risk, value of information, and decisions under uncertainty. T E
Multi-agent systems and game theory Cooperation, competition, equilibria, basic mechanism design, and decentralized decision making. T E

6. Classical Statistical Learning

Topic Expected competency Category Level
Problem formulation Regression, classification, ranking, clustering, density estimation, anomaly detection, and structured prediction. B C
Linear regression Least squares, feature transformations, residual analysis, regularization, and interpretation. B C
Logistic regression Log-odds, cross-entropy, decision thresholds, multiclass extensions, and calibration. B C
Regularization L1 and L2 penalties, sparsity, weight decay, model capacity, and the bias-variance trade-off. B C
K-nearest neighbors Distance metrics, scaling, neighborhood size, classification/regression, and computational trade-offs. B C
Decision trees Splitting criteria, pruning, depth control, missing values, and interpretability. B C
Ensembles Bagging, random forests, gradient boosting, stacking, and when ensembles improve robustness. B C
Support vector machines and kernels Margins, soft-margin SVMs, kernel functions, and scaling considerations. B C
Probabilistic classifiers Naive Bayes, linear/quadratic discriminant analysis, posterior probabilities, and calibration. B E
Model evaluation Accuracy, precision, recall, F1, log loss, ROC-AUC, PR-AUC, ranking metrics, regression errors, and task-specific metrics. B C
Class imbalance and cost-sensitive learning Resampling, class weights, threshold selection, focal-style objectives, and asymmetric error costs. B C
Clustering K-means, hierarchical clustering, DBSCAN, spectral clustering, assumptions, and cluster validation. B C
Dimensionality reduction PCA, t-SNE, UMAP, feature selection, visualization, and limits of low-dimensional projections. B C
Semi-supervised and weakly supervised learning Pseudo-labeling, consistency ideas, noisy labels, and use of unlabeled data. B E
Anomaly and novelty detection One-class methods, isolation-based methods, reconstruction error, and thresholding. B E
Domain shift and generalization Covariate shift, label shift, domain adaptation, invariance, and robust validation. B E

7. Neural Networks and Deep Learning

Topic Expected competency Category Level
Perceptrons and multilayer networks Perceptron basics, multilayer perceptrons, universal approximation intuition, and network capacity. B C
Activations and losses ReLU, sigmoid, tanh, softmax, MSE, MAE, cross-entropy, margin losses, and task-appropriate objectives. B C
Backpropagation Forward and backward passes, computational graphs, gradient flow, and common implementation errors. B C
Optimization for deep learning Apply the optimization methods in Section 3 to neural networks; learning-rate warmup and gradient clipping. B C
Initialization and normalization Weight initialization, batch normalization, layer normalization, and their effects on training. B C
Regularization Dropout, early stopping, weight decay, augmentation, label smoothing, and model selection. B C
Embeddings and pooling Learned representations for text, images, audio, categories, and graphs; max, average, and attention pooling. B C
Convolutional networks Convolution, stride, padding, receptive fields, pooling, residual connections, and image/time-series use. B C
Recurrent sequence models RNNs, LSTMs, GRUs, hidden state, sequence batching, teacher forcing, and vanishing/exploding gradients. B C
Attention and transformers Attention mechanisms, self-attention, encoder/decoder blocks, masking, and positional information. B C
Autoencoders Undercomplete, denoising, and variational autoencoders for representation learning and generation. B C
Transfer learning and fine-tuning Pretrained encoders, feature extraction, full fine-tuning, parameter-efficient fine-tuning, and catastrophic forgetting. B C
Self-supervised and contrastive learning Pretext tasks, augmentations, positive/negative pairs, contrastive objectives, and representation evaluation. B E
Training diagnosis Learning curves, gradient checks, dead activations, unstable loss, overfitting, underfitting, and systematic ablations. P C

8. Reinforcement Learning and Sequential Decision Making

Topic Expected competency Category Level
Multi-armed bandits Exploration-exploitation trade-off, epsilon-greedy methods, upper-confidence ideas, and regret intuition. B C
Markov decision processes States, actions, transitions, rewards, policies, returns, discounting, and episodic/continuing tasks. B C
Value functions and Bellman equations State/action values, Bellman expectation and optimality equations, and fixed-point intuition. B C
Dynamic programming Policy evaluation, policy iteration, and value iteration when a model is known. B C
Monte Carlo and temporal-difference learning Prediction from experience, bootstrapping, TD error, n-step returns, and eligibility-trace intuition. B C
Control methods SARSA, Q-learning, on-policy vs off-policy learning, and exploration strategies. B C
Function approximation and deep RL Value approximation, DQN, replay buffers, target networks, and instability sources. B E
Policy-gradient and actor-critic methods REINFORCE, baselines, advantages, actor-critic structure, entropy regularization, and PPO. B E
Reward design and curriculum learning Sparse rewards, shaping, rewards for satisfying constraints and penalties for violations, staged objectives, and unintended incentives. B E
Imitation, offline, and model-based RL Behavior cloning, offline data limitations, learned dynamics, planning with models, and sim-to-real intuition. B E
RL evaluation and reproducibility Apply evaluation and reproducibility methods from Section 4; multiple seeds, return distributions, sample efficiency, and stability. B C
Multi-agent reinforcement learning Cooperative and competitive settings, non-stationarity, centralized training, and decentralized execution. B E

9. Computer Vision

Topic Expected competency Category Level
Image fundamentals Pixels, channels, color spaces, resizing, interpolation, normalization, and common image formats. B C
Convolutional layers Apply the convolutional networks in Section 7 to images; kernels and feature maps. B C
Image classification Training classifiers, transfer learning, pretrained encoders such as ResNet, and error analysis. P C
Object detection Bounding boxes, IoU, non-maximum suppression, and practical use of YOLO, SSD, and DETR-style models. B C
Image segmentation Semantic/instance segmentation, pixel-wise losses, U-Net-style models, and overlap metrics. B C
Image augmentation Cropping, flipping, geometric and photometric transforms, mix-based augmentations, and label consistency. P C
Vision-language encoders Joint image-text embeddings, CLIP-style similarity, zero-shot classification, and retrieval. B C
Generative vision GANs, variational methods, diffusion models, conditioning, sampling, and evaluation limitations. B C
Video and temporal vision Frame sampling, temporal pooling, motion features, sequence models, and video classification. B E

10. Natural Language Processing, Speech, and Audio

Topic Expected competency Category Level
Text preprocessing and tokenization Apply data processing from Section 4 to text; sentence/word/subword tokenization, vocabularies, and sequence length. B C
Text representations Bag-of-words, TF-IDF, static embeddings, contextual embeddings, and similarity measures. B C
Text classification and sequence labeling Document classification, token classification, class imbalance, and evaluation. P C
Pretrained text encoders BERT-style models, masked-language-model representations, pooling, and fine-tuning. B C
Language modeling Autoregressive and masked objectives, likelihood, perplexity, generation, and decoding. B C
Encoder-decoder models Sequence-to-sequence learning, attention, machine translation, summarization, and multimodal generation. B C
Retrieval and semantic search Dense and sparse retrieval, embeddings, similarity, reranking, and retrieval evaluation. B E
Audio and signal representations Waveforms, sampling rate, resampling, framing, spectrograms, Fourier transforms, and normalization. B C
Speech and audio models Practical use of models such as Whisper, Qwen-Audio, and Voxtral when provided in the contest environment. P C

11. Transformers, Foundation Models, and Large Language Models

Topic Expected competency Category Level
Transformer mechanics Scaled dot-product attention, multi-head attention, masking, positional encoding, residual connections, and normalization. B C
Architecture families Encoder-only, decoder-only, encoder-decoder, mixture-of-experts intuition, and modality-specific transformers. T C
Tokenization and context Apply the tokenization methods in Section 10 to foundation models; context windows, truncation, attention masks, and prompt construction. B C
Pretraining objectives Next-token prediction, masked modeling, denoising, instruction tuning, and their behavioral consequences. T C
Prompt engineering Zero-shot/few-shot prompting, role and instruction design, examples, decomposition, context selection, and prompt evaluation. B C
Structured generation and decoding Greedy, beam, sampling, temperature/top-p, constrained outputs, validation, repair, and stopping criteria. B C
Fine-tuning and PEFT Supervised fine-tuning, LoRA/QLoRA, adapters, prompt tuning, dataset construction, and hyperparameter sensitivity. B C
Preference and reinforcement-based adaptation Reward models, RLHF, DPO-style objectives, PPO/GRPO-style optimization, and alignment trade-offs. B E
Retrieval-augmented generation Chunking, embeddings, vector search, retrieval, reranking, grounding, and citation-aware evaluation. B E
Tool use and agents Function/tool calling, planning loops, memory, state tracking, environment feedback, and rule-constrained actions. B E
LLM evaluation Task accuracy, exact match, validity, calibration, hallucination, robustness, preference evaluation, and contamination risks. B C
Inference efficiency Quantization, batching, caching, sequence length, memory/latency trade-offs, and small-model selection. B E
Multimodal foundation models Joint text-image-audio inputs, modality encoders, fusion, prompting, and cross-modal evaluation. B E
LLM safety and security Prompt injection, jailbreaks, data leakage, unsafe outputs, tool misuse, and defensive validation. B C
Use of APIs and provided models Use models and APIs provided for the task. P C

12. Learning from Time Series, Sequences, and Sensor Data

Topic Expected competency Category Level
Sampling and segmentation Sampling frequency, windows, overlap, resampling, synchronization, variable-length sequences, and event segmentation. B C
Preprocessing Apply data processing from Section 4 to time series; filtering, detrending, outliers, and leakage-safe transformations. B C
Statistical and frequency-domain features Moments, extrema, autocorrelation, spectral energy, dominant frequencies, and Fourier-based features. B C
Classical forecasting Autoregression, moving averages, seasonality, trend, baseline forecasts, and ARIMA-style intuition. B E
Neural sequence models RNN/LSTM/GRU, temporal CNNs, transformers, sequence-to-one, sequence-to-sequence, and multivariate inputs. B C
Multi-step forecasting Direct and autoregressive prediction, teacher forcing, rollout, exposure bias, error accumulation, and horizon selection. B C
Temporal validation and metrics Chronological splits, rolling evaluation, event-based splits, RMSE/MAE, scale-normalized metrics, and leakage control. B C
Anomaly and change-point detection Point/contextual anomalies, reconstruction/prediction errors, change points, and threshold selection. B E
Sequence similarity and alignment Cross-correlation, dynamic time warping, learned similarity, and invariance to phase or speed. B E
Learning from sensor and IMU data Accelerometer/gyroscope data, axes, orientation effects, sensor fusion, activity recognition, and biometric verification. B E
Curriculum and long-horizon training Progressively increasing sequence difficulty or forecast horizon while monitoring forgetting and stability. B E

13. Graph Machine Learning and Geometric Deep Learning

Topic Expected competency Category Level
Graph representations Nodes, edges, attributes, directed/undirected graphs, adjacency matrices, edge lists, sparse tensors, and batching. B C
Classical graph features Degree, centrality, neighborhoods, paths, connected components, Laplacian intuition, and handcrafted graph statistics. B E
Message passing Neighborhood aggregation, update functions, receptive fields, permutation invariance, and oversmoothing intuition. B C
GNN architectures GCN, GraphSAGE, GAT, edge-aware message passing, residual connections, and normalization. B C
Graph learning tasks Node classification/regression, link prediction, edge prediction, graph classification/regression, and graph embeddings. B C
Pooling and readout Node-to-graph aggregation, hierarchical pooling intuition, and global representations. B E
Graph construction Building edges from physical relations, similarity, proximity, domain knowledge, and learned connectivity. B C
Temporal and spatio-temporal graphs Dynamic node features, temporal message passing, recurrent/temporal GNNs, and spatio-temporal forecasting. B E
Heterogeneous and dynamic graphs Multiple node/edge types, time-varying topology, relation-specific parameters, and practical representations. B E
Scalability and implementation Sparse operations, neighbor sampling, mini-batching, memory constraints, and graph data loaders. P E
Domain and physical constraints Directional flow, conservation, topology-aware losses, and integrating scientific priors into graph models. B E

14. Contest Tasks

What to Expect

  • Tasks involve implementing solutions, training or using models, producing predictions, or choosing actions in interactive environments.
  • A task may combine tabular data, images, text, audio, video, time series, sensor streams, or graphs.
  • Solving a task involves adapting methods to its data and constraints, testing ideas, and using the results to improve the solution within the time and compute budget.
  • Solutions that improve on the baseline can earn partial credit. Stronger solutions may combine several insights.

What Not to Expect

  • Solving a task simply by applying an off-the-shelf model or method unchanged.
  • Questions that depend on memorizing obscure library APIs.

These resources support preparation. The ICAIC topic tables above define the examinable scope; the recommended reading does not add requirements.

Resource Subject
Artificial Intelligence: A Modern Approach — Stuart Russell and Peter Norvig Classical AI.
The Elements of Statistical Learning — Trevor Hastie, Robert Tibshirani, and Jerome Friedman Statistical learning.
Reinforcement Learning: An Introduction — Richard S. Sutton and Andrew G. Barto Reinforcement learning.
Deep Learning — Ian Goodfellow, Yoshua Bengio, and Aaron Courville Deep learning.

Supplementary Resources

This site uses Just the Docs, a documentation theme for Jekyll.