Revision
University of Essex - CE213 Introduction to Artificial Intelligence
Learning Outcomes
Explain and criticize arguments for and against the possibility of artificial intelligence.
Implement standard blind and heuristic search procedures; understand their strengths and weaknesses.
Explain operation of production system interpreters and their pros and cons.
Understand established machine learning procedures and appropriate problem types.
Understand agent-oriented AI and build multi-agent systems using subsumption architecture.
Basic AI Concepts and Methods
Key Terms
AI Machine Learning: Knowledge Representation, Search
Critical Concepts: Generate and Evaluate, State Space Representation
Search Strategies: Blind Search, Heuristic Search, Game Playing
Methods
Importance of Knowledge Production System
Forward Chaining
Backward Chaining
MYCIN as a production system example
Problem Solving Skills - Checklist
Formalize problems into state space representation.
Analyze properties of search strategies (4 criteria).
Find optimal routes using search strategies and heuristics.
Minimax search for game trees using heuristic values.
Update game trees through MCTS iterations.
Identify rule firing and conclusions using forward chaining.
Determine certainty of conclusions using backward chaining and MYCIN’s system.
Calculate information and information gain using Shannon’s formula.
Induce decision trees based on information gain.
Sample Questions and Revision Material
Review CE213 exam rubric and sample questions on Moodle.
Past CE213 exam papers available for practice.
Note: Exam format is Open Book (Restricted), no bookwork questions.
Typical applications: puzzles, game playing, robot control, prediction, data mining.
Useful Equations
CF_combined = CFr1 + (1 - CFr1) x CFr2
P(X|Y) = P(Y|X) x P(X) / P(Y)
Information = -3p5log8(p5)
Academic Support
Academic Support Hours for CE213 revision: Thursdays 4-6pm on Zoom.
CE213 Help Forum on Moodle for queries: Post discussion topics for assistance.
Lecture Summaries
Lecture 1: What is AI?
AI defined as science and engineering of making intelligent machines.
Discussion on the feasibility of AI including Lady Lovelace's objection, Turing Test, and Searle's Chinese Room.
Lecture 2: Problem Solving
Example: Corn, Goose and Fox Puzzle - essence of problem abstraction and systematic search.
Lecture 3: Search Strategies
Types: Breadth-First Search, Depth-First Search, Iterative Deepening, Uniform Cost Search.
Comparison based on completeness, optimality, time, and space complexity.
Lecture 4: Advanced Search Strategies
Greedy Search and A* Search - optimal and efficient but need good heuristics.
Overview of Hill Climbing.
Lectures 5 & 6: Minimaxing & Adversarial Search
Evaluation functions for adversarial search, alpha-beta pruning implementation, Monte-Carlo Tree Search overview.
Lecture 7: Knowledge and Problem Solving
Production systems for knowledge representation; procedural representation.
Lecture 8: Production Systems
Core components: production rules, interpreters, environment.
Lecture 9: MYCIN System
Knowledge representation methods; reasoning with uncertainty and explanation generation.
Lecture 10: Machine Learning Fundamentals
Key elements, functioning, and tasks involved in machine learning.
Lecture 11: Decision Trees
Structure, induction procedures, and information gain evaluation.
Lecture 12: Challenges in Decision Trees
Issues like inconsistent data, numeric attributes, and overfitting.
Lectures 13-14: Neural Networks
MP Neurons, learning rules, delta rule limitations, and necessity of nonlinear neurons for complex models.
Lectures 15: Error Back-Propagation
Understanding error back-propagation for multilayer neural networks and its challenges.
Lecture 16: Clustering Techniques
Methods: Agglomerative Hierarchical Clustering, K-Means; evaluation metrics.
Lectures 17-18: Reinforcement Learning & Q Learning
Concepts, representation of MDPs, Q learning algorithms, and usage in practice.
Lectures 19-20: Genetic Algorithms & Agents
Basis of genetic algorithms; characteristics of intelligent agents, architecture arrangements for task accomplishment.