|Delivery Type||Delivery length / details|
|Assessment Type||Assessment length / details||Proportion|
|Semester Exam||2 Hours Written Exam||100%|
|Supplementary Exam||2 Hours Resit failed examination||100%|
On successful completion of this module, students will be able to:
1. Formalise and represent experience, using techniques like feature-value pairs;
2. Apply concept learning, decision making tree, artificial neural networks, Bayesian learning, genetic algorithms, and reinforcement learning to solve a range of simple problems in applications like customer attitude prediction, banking business, handwritten character recognition, medical diagnosis, artificial life, robot path optimization, etc.;
3. Compare and choose different learning algorithms for different applications;
4. Appreciate the range of applicability of intelligent learning concepts and techniques; ;
5. Explain the state of the art of intelligent learning concepts and techniques;
This module explores real world applications of machine learning. To this end, a small number of topics are studied in depth in order to give insight and understanding of the methods and issues involved in the state-of-the-art applications of the various learning techniques.
Possibility and necessity of learning, target function, components in a learning system, performance measurement of learning systems.
2. Concept learning - 3 lectures
This chapter will use concept learning to develop applications to predict whether students like this module or not: Generality ordering of hypotheses, FIND-S algorithm, version space, the LIST-THEN-ELIMINATE algorithm, inductive bias
3. Decision tree learning - 3 lectures
This chapter will use decision tree learning to develop applications about banking business: Entropy, best attribute, information gain, best tree, inductive bias, Occam's razor, over-fitting, reduced error pruning, rule post-pruning.
4. Artificial neural network - 3 lectures
This chapter will use artificial neural network learning to develop applications to recognize handwritten characters: Perceptron, linear separability, gradient decent, sigmoid function, back propagation algorithm, over-fitting.
5. Bayesian learning - 4 lectures
This chapter will use Bayesian learning to develop applications about medical diagnosis: Bayesian theory, maximum a posteriori hypothesis, maximum likelihood, probability density, normal distribution, minimum description length principle, Bayes optimal classifier, naive Bayes classifier.
6. Support vector machines - 3 lectures
This chapter will introduce the ideas, formulations and applications of support vector machines: Binary linear classifier, maximum margin hyperplane, linear separability/non-linear separability, kernel function, hard/soft margin, primal/dual form, quadratic programming, cross validation, applications.
7. Reinforcement learning - 2 lectures
This chapter will use reinforcement learning to develop applications about robot path optimization: Reward, Markov decision process, utility function, Q-learning.
Reading ListRecommended Text
S J Russell and P Norvig (1995) A I: A Modern Approach Prentice-Hall Primo search S. M. Weiss and C. A. Kulikowsky (1991) Computer Systems That Learn Morgan Kaufmann Primo search T. Mitchell (1998) Machine Learning McGraw Hill Primo search
This module is at CQFW Level 6