Artificial intelligence Course for Computer Science

Lesson 1 • Heuristic and Informed Search

Introduces A*, greedy best-first, and admissible heuristic design. Connects search efficiency to domain-specific knowledge encoding.

Lesson 2 • Problem Formulation and State Spaces

Formalises problems as state-space graphs with goals and operators. Enables learners to translate real tasks into solvable computational models.

Lesson 3 • History and Scope of AI

Traces AI from symbolic reasoning to modern deep learning milestones. Provides context for evaluating current techniques against foundational goals.

Lesson 4 • Intelligent Agents and Environments

Defines agents, rationality, and environment types using the PEAS framework. Grounds subsequent algorithm design in agent-based thinking.

Lesson 1 • First-Order Predicate Logic

Extends propositional logic with quantifiers, predicates, and functions. Enables richer knowledge representation for real-world domains.

Lesson 2 • Non-Monotonic and Temporal Reasoning

Addresses default logic, closed-world assumption, and temporal logic. Equips learners to handle incomplete and time-varying knowledge.

Lesson 3 • Propositional Logic for AI

Covers syntax, semantics, and inference rules of propositional logic. Establishes the formal basis for all subsequent reasoning systems.

Lesson 4 • Reasoning Under Uncertainty

Covers probability theory, Bayes' theorem, and belief updating. Prepares learners for probabilistic AI models in later chapters.

Lesson 1 • Decision Trees and Ensemble Methods

Covers ID3, CART, random forests, and gradient boosting. Demonstrates how ensemble strategies reduce variance and improve generalisation.

Lesson 2 • Unsupervised Learning Methods

Covers k-means, hierarchical clustering, PCA, and autoencoders. Equips learners to discover structure in unlabelled data sets.

Lesson 3 • Model Evaluation and Selection

Teaches confusion matrices, ROC curves, and hyperparameter tuning. Ensures learners can rigorously compare and select models.

Lesson 4 • Learning Paradigms and Workflow

Defines supervised, unsupervised, and reinforcement learning with the standard ML pipeline. Frames all subsequent model-specific content.

Lesson 5 • Linear and Logistic Regression

Derives least-squares regression and logistic classification from first principles. Provides the mathematical foundation for gradient-based learning.

Lesson 1 • Convolutional Neural Networks

Explains convolution, pooling, and standard CNN architectures for vision. Prepares learners for image-based AI applications.

Lesson 2 • Backpropagation and Optimisation

Derives backpropagation via chain rule and covers modern optimisers. Enables learners to diagnose and fix training instability.

Lesson 3 • Regularisation and Generalisation

Covers dropout, batch normalisation, weight decay, and early stopping. Directly addresses overfitting in deep models.

Lesson 4 • Perceptrons and Feedforward Networks

Derives the perceptron learning rule and multilayer feedforward architecture. Establishes the computational graph view used throughout deep learning.

Lesson 1 • Attention Mechanisms and Transformers

Derives scaled dot-product attention and the transformer architecture. Provides the theoretical basis for large language models.

Lesson 2 • Sequence Labelling and Parsing

Covers POS tagging, NER, and dependency parsing with CRFs and neural models. Builds structural understanding of language for information extraction.

Lesson 3 • Pre-trained Language Models

Covers BERT, GPT-style models, and fine-tuning strategies. Enables learners to adapt large models to domain-specific NLP tasks.

Lesson 4 • Text Preprocessing and Representation

Covers tokenisation, stemming, TF-IDF, and bag-of-words encoding. Establishes the data pipeline foundation for all NLP models.

Lesson 1 • Image Representation and Preprocessing

Covers pixel arrays, colour spaces, normalisation, and augmentation strategies. Establishes the data foundation for all vision models.

Lesson 2 • Semantic and Instance Segmentation

Covers FCN, U-Net, and Mask R-CNN for pixel-level prediction. Extends detection skills to fine-grained scene understanding.

Lesson 3 • Generative Models for Images

Introduces VAEs and GANs for image synthesis and data augmentation. Expands learners' toolkit to generative AI applications.

Lesson 4 • Vision Transformers and Modern Architectures

Covers ViT, CLIP, and self-supervised vision pre-training. Connects NLP transformer knowledge to state-of-the-art vision models.

Lesson 1 • Policy Gradient Methods

Derives REINFORCE and actor-critic algorithms for continuous action spaces. Extends RL beyond discrete action settings.

Lesson 2 • Markov Decision Processes

Formalises sequential decision-making with states, actions, rewards, and policies. Provides the mathematical framework for all RL algorithms.

Lesson 3 • Advanced RL Algorithms

Covers PPO, SAC, and model-based RL for sample efficiency. Prepares learners for state-of-the-art RL research and applications.

Lesson 4 • Temporal Difference Learning

Covers TD(0), SARSA, and Q-learning with convergence guarantees. Bridges dynamic programming to model-free online learning.

Lesson 1 • Serving and Monitoring in Production

Addresses REST APIs, batch inference, model drift detection, and alerting. Ensures deployed models remain accurate and reliable over time.

Lesson 2 • Model Compression and Optimisation

Covers pruning, quantisation, knowledge distillation, and ONNX export. Prepares models for deployment on resource-constrained hardware.

Lesson 3 • Scalable Training Infrastructure

Teaches data parallelism, model parallelism, and distributed training frameworks. Enables training of large models on multi-device clusters.

Lesson 4 • MLOps and Model Lifecycle Management

Covers experiment tracking, model versioning, CI/CD for ML, and feature stores. Connects research prototypes to reliable production systems.