PolyU | Artificial Intelligence Concepts

📚 Content Summary

This comprehensive course provides a rigorous yet accessible introduction to Artificial Intelligence, designed for postgraduate students and professionals. Bridging the gap between historical foundations and cutting-edge innovations, the curriculum progresses from symbolic AI and search algorithms to modern Deep Learning and Generative AI. Students will explore essential topics such as knowledge representation, probabilistic reasoning, and classical machine learning before diving deep into neural networks, Transformers, and Large Language Models (LLMs). Emphasizing both theory and practice, the course utilizes Python and industry-standard frameworks like PyTorch to implement algorithms, interact with modern APIs, and address critical issues in AI ethics and safety.

Bridging the gap between classical Symbolic AI and modern Generative AI, this course equips students with the theoretical depth and practical Python skills to architect, implement, and ethically evaluate intelligent systems.

🎯 Learning Objectives

1. Master AI Foundations: Analyze and implement fundamental problem-solving paradigms, including heuristic search, logic-based reasoning, and probabilistic modeling.
2. Implement Deep Learning Architectures: Design and train advanced neural networks using PyTorch, ranging from Multi-layer Perceptrons to Convolutional Neural Networks and Transformers.
3. Deploy Generative AI Solutions: Engineer applications utilizing Large Language Models (LLMs), employing techniques such as fine-tuning, prompt engineering, and Retrieval-Augmented Generation (RAG).
4. Ensure Ethical AI Development: Critically evaluate AI systems for bias, safety, and alignment, applying strategies for explainability and robustness in real-world scenarios.

---

🔹 Lesson 1: History and Foundations of AI

Overview: An exploration of the origins of Artificial Intelligence, covering the Turing Test, the Dartmouth Conference, and the cyclical nature of AI development known as 'AI Winters'.

Learning Outcomes:

• Trace the historical evolution of AI from 1956 to the present day.
• Compare and contrast Symbolic AI (GOFAI) and Connectionism paradigms.
• Discuss the impact of the Deep Learning Revolution and recent generative breakthroughs.

---

🔹 Lesson 2: Search Strategies: Uninformed and Informed

Overview: Introduction to fundamental problem-solving agents using search algorithms, distinguishing between blind search and heuristic-based approaches.

Learning Outcomes:

• Implement uninformed search methods including BFS, DFS, and Uniform Cost search.
• Apply heuristic search techniques such as A* and Greedy Best-First search.
• Analyze the complexity and optimality of different search strategies.

---

🔹 Lesson 3: Adversarial Search and Constraint Satisfaction

Overview: A deep dive into decision-making in competitive environments (games) and solving problems defined by constraints.

Learning Outcomes:

• Master the Minimax algorithm and Alpha-Beta pruning for game playing.
• Understand Monte Carlo Tree Search (MCTS) used in advanced systems like AlphaGo.
• Solve Constraint Satisfaction Problems (CSP) using backtracking and local search.

---

🔹 Lesson 4: Knowledge Representation and Reasoning

Overview: Study of how AI systems represent information logically to perform inference, moving from classical logic to modern knowledge structures.

Learning Outcomes:

• Utilize Propositional and First-Order Logic for representation and inference.
• Compare legacy Expert Systems with modern Knowledge Graphs and Ontologies.
• Explore Neuro-Symbolic AI as a bridge between logic and neural networks.

---

🔹 Lesson 5: Uncertainty Management

Overview: Techniques for reasoning in environments where information is incomplete or probabilistic.

Learning Outcomes:

• Apply Bayesian probability theory to AI decision-making.
• Construct and analyze Bayesian Networks and Markov Decision Processes (MDPs).
• Understand temporal probability models through Hidden Markov Models (HMMs).

---

🔹 Lesson 6: Classical Machine Learning

Overview: A practical introduction to data-driven algorithms for supervised and unsupervised learning using Python.

Learning Outcomes:

• Implement supervised learning models including Decision Trees and SVMs.
• Apply unsupervised techniques like K-means clustering and PCA.
• Use Scikit-learn to solve basic data analytic problems.

---

🔹 Lesson 7: Neural Network Foundations

Overview: The transition to biological-inspired computing, covering the architecture and training mechanisms of artificial neural networks.

Learning Outcomes:

• Explain the structure of artificial neurons and Multi-layer Perceptrons (MLP).
• Master the Backpropagation algorithm and various activation functions (ReLU, GELU).
• Build a basic neural network using PyTorch.

---

🔹 Lesson 8: Computer Vision and CNNs

Overview: Focus on processing visual data through specialized deep learning architectures.

Learning Outcomes:

• Understand the mechanics of Convolutional Neural Networks (CNNs).
• Analyze advanced architectures like ResNet and Vision Transformers (ViT).
• Implement a basic image classification model.

---

🔹 Lesson 9: Sequence Modeling and The Transformer

Overview: Evolution of sequence processing from Recurrent Neural Networks to the revolutionary Attention mechanism.

Learning Outcomes:

• Differentiate between RNNs, LSTMs, and the Transformer architecture.
• Explain the 'Attention Is All You Need' paper and self-attention mechanisms.
• Understand the role of Encoders and Decoders in models like BERT and GPT.

---

🔹 Lesson 10: Large Language Models (LLMs)

Overview: In-depth study of state-of-the-art Generative AI focusing on text generation and model adaptation.

Learning Outcomes:

• Describe the Pre-training and Fine-tuning lifecycle of LLMs.
• Explore Reinforcement Learning from Human Feedback (RLHF).
• Practice Prompt Engineering and parameter-efficient fine-tuning methods (PEFT/LoRA).

---

🔹 Lesson 11: Advanced Generative AI Applications

Overview: Expanding beyond text to retrieval systems and generative media using modern APIs.

Learning Outcomes:

• Implement Retrieval-Augmented Generation (RAG) to reduce hallucinations.
• Understand the math behind Diffusion Models (e.g., Stable Diffusion) and VAEs.
• Interact with modern AI APIs to build a generative application.

---

🔹 Lesson 12: AI Ethics, Safety, and Alignment

Overview: Critical analysis of the societal impact of AI, focusing on responsibility and safety protocols.

Learning Outcomes:

• Identify sources of bias and ensure fairness in AI models.
• Discuss AI Safety, Robustness, and the Alignment problem.
• Evaluate Explainability (XAI) and mitigation strategies for LLM hallucinations.

---