Artificial Intelligence (AI) Course

Lesson 1 • AI Ethics and Responsible Design

Introduces bias, fairness, transparency, and accountability principles. Establishes ethical reasoning as a non-negotiable design constraint throughout the course.

Lesson 2 • Knowledge Representation Basics

Explains propositional and first‑order logic plus semantic networks, showing how symbolic representations encode facts and relationships. Shows integration combining symbolic and neural methods.

Lesson 3 • Intelligent Agents and Environments

Defines rational agents, PEAS framework, and environment types. Students can design agent architectures suited to specific task environments.

Lesson 4 • History and Evolution of AI

Traces AI from symbolic logic to deep learning, highlighting pivotal breakthroughs. Provides context for understanding why modern techniques emerged and where they excel.

Lesson 1 • Supervised Learning Fundamentals

Covers labeled data, loss functions, and gradient-based optimisation. Forms the backbone for understanding neural networks and advanced models later.

Lesson 2 • Classification Algorithms in Depth

Examines logistic regression, decision trees, SVMs, and k‑NN. Covers training pipelines, hyperparameters, and evaluation so students can implement and compare each classifier on benchmark datasets.

Lesson 3 • Regression and Prediction Models

Builds linear and polynomial regression models with statistical interpretation. Connects prediction accuracy to business and scientific decision-making.

Lesson 4 • Unsupervised Learning Methods

Explores clustering, dimensionality reduction, and density estimation. Students discover hidden structure in unlabeled data for exploratory analysis.

Lesson 1 • Backpropagation and Optimisation

Derives backpropagation via chain rule and covers Adam, RMSProp, and SGD. Students debug vanishing gradients and tune optimisers effectively.

Lesson 2 • Recurrent Networks and Sequence Models

Examines recurrent neural networks, LSTM and GRU units for modelling sequential data. Covers unrolling, back‑propagation through time, and practical considerations, linking these techniques to NLP and time‑series forecasting.

Lesson 3 • Regularisation and Practical Training

Applies dropout, early stopping, and mixed-precision training to real networks. Students reduce overfitting and cut training time on GPU hardware.

Lesson 4 • Convolutional Neural Networks

Covers convolution, pooling, and feature map interpretation for image tasks. Students build CNNs and apply transfer learning to custom datasets.

Lesson 1 • Word Embeddings and Semantic Space

Trains and applies Word2Vec, GloVe, and FastText embeddings, showing how dense vectors capture semantic similarity. Includes evaluation techniques and visualisations for exploring embedding spaces.

Lesson 2 • Transformer Architecture Deep Dive

Dissects multi-head self-attention, positional encoding, and encoder-decoder stacks. Provides the architectural understanding needed to fine-tune large language models.

Lesson 3 • Pre-trained Models and Fine-Tuning

Fine-tunes BERT, RoBERTa, and GPT variants on domain-specific tasks. Students achieve state-of-the-art NLP results with limited labeled data.

Lesson 4 • NLP Applications and Pipelines

Builds end-to-end pipelines for sentiment analysis, NER, and summarisation. Integrates preprocessing, modelling, and deployment into production-ready systems.

Lesson 1 • Vision Transformers and Modern Architectures

Applies Vision Transformers, CLIP, and DINO for image classification and zero‑shot recognition. Explains patch embedding, self‑attention, and contrastive learning, linking NLP transformer concepts to modern vision systems.

Lesson 2 • Generative Models for Images

Builds GANs and variational autoencoders to synthesise realistic images and perform style transfer. Explores latent space manipulation, training stability tricks, and metrics for assessing generation quality and diversity.

Lesson 3 • Object Detection Architectures

Compares YOLO, Faster R-CNN, and SSD for real-time and accurate detection. Students train detectors on custom datasets with bounding box annotations.

Lesson 4 • Semantic and Instance Segmentation

Implements U‑Net and Mask R‑CNN for pixel‑level segmentation, detailing encoder‑decoder designs and mask generation pipelines. Connects these techniques to domains like medical imaging, autonomous driving, and robotics.

Lesson 1 • Model-Free Value-Based Methods

Implements Q-learning, SARSA, and Deep Q-Networks on Gym environments. Students understand temporal difference learning and experience replay.

Lesson 2 • Markov Decision Processes

Formalises states, actions, rewards, and transition dynamics mathematically. Provides the theoretical foundation for all reinforcement learning algorithms.

Lesson 3 • Multi-Agent and Hierarchical RL

Extends RL to cooperative and competitive multi-agent settings and skill hierarchies. Prepares students for real-world complex system control.

Lesson 4 • Advanced RL: PPO and SAC

Applies Proximal Policy Optimisation and Soft Actor-Critic to complex tasks. Students achieve stable training on continuous control benchmarks.

Lesson 1 • Fine-Tuning and Alignment

Applies supervised fine‑tuning, RLHF, and direct preference optimisation to align LLMs with human intent. Students create domain‑specific models that follow instructions while adhering to safety constraints.

Lesson 2 • Prompt Engineering and In-Context Learning

Applies zero-shot, few-shot, chain-of-thought, and structured prompting techniques. Effective prompting multiplies model capability without any parameter updates.

Lesson 3 • LLM Architecture and Scaling Laws

Analyses GPT, LLaMA, and Mistral architectures alongside Chinchilla scaling laws. Students predict model capability from compute and data budgets.

Lesson 4 • LLM Agents and Tool Use

Builds autonomous agents using ReAct, function calling, and multi-agent frameworks. Students deploy agents that plan, use tools, and complete multi-step tasks.

Lesson 1 • Monitoring and Model Maintenance

Implements drift detection, alerting, and retraining triggers in production. Prevents silent model degradation that erodes business value over time.

Lesson 2 • Experiment Tracking and Reproducibility

Uses MLflow, Weights & Biases, and DVC to log experiments systematically. Reproducibility enables team collaboration and regulatory compliance.

Lesson 3 • Model Serving and Deployment

Deploys models via REST APIs, batch pipelines, and edge runtimes. Guides students through Docker containerisation, model serialisation, and scalable serving strategies for performance and reliability.

Lesson 4 • Scalable Training Infrastructure

Configures distributed training with Horovod and cloud GPU clusters. Students reduce training time for large models through parallelism strategies.