Aims

Learning outcomes

• be aware of the implications of a problem domain for the selection of knowledge acquisition methodology, knowledge engineering technique, software development methodology, and system evaluation strategy;

• understand the differences between the main classes of application oriented intelligent system, including representation and reasoning facilities, and performance issues;

• understand the different techniques and methodologies available for knowledge acquisition, and their relative strengths and weaknesses;

• appreciate the problem of software engineering for intelligent systems;

• have had practical exposure to a range of intelligent systems: including an expert system shell; a CBR toolkit; a neural network simulator and a knowledge acquisition environment.

General description

Syllabus

Typical Knowledge Based System Applications
The different types of knowledge-based system (KBS) application and the techniques that have proved to be useful.

Knowledge Acquisition
These lectures will consider the issue of obtaining the knowledge to be used in an expert system. They will be based around the KADS formalism which will be briefly introduced. This formalism guides the expert system building through the process of task identification and knowledge acquisition.

Expert Systems
This section will build on the introduction to expert systems presented earlier. It will introduce the students to commercial expert system shells and will provide hands-on experience of at least one system. It will consider the issues associated with constructing expert systems and of rule-based representations.

Case Based Reasoning
Case-based reasoning (CBR) systems: the basic structure of a CBR system and the use of a case base, representation and indexing of cases, case retrieval and adaptation, and case repair. Strengths and weaknesses of CBR: when to apply and when not to apply it.

Neural Networks
The concept behind neural networks: the more important and useful neural network architectures. The application of neural computing in industry.

Genetic Algorithms and Classifier Systems
What is meant by a genetic algorithm or classifier system and how it is constructed. Theoretical aspects of genetic algorithms such as building blocks, and the factors which make a problem hard for a genetic algorithm. The lectures will introduce the various issues in evolutionary computing (such as representation, evaluation, selection, mutation and crossover) as well as available toolkits. To accompany these lectures practical sessions will be run where students will carry out various assessed assignments.

Machine Induction
There are a large number of applications which exploit the approach to machine learning known as induction. In particular a number of case-based systems and data mining applications have made machine induction commercially useful (and viable). In these lectures the ID3 algorithm, and its enhancements will be described and the commercial uses of these algorithms discussed.

Model-Based Reasoning
In this section of the course, functional model-based reasoning will be introduced. How a functional model-based system can be constructed and the problems and issues associated with such a task will be considered. In addition a functional model-based system constructed for engineers will be considered in detail. This case study will highlight the trade-offs which have to be made between ease of use and flexibility/power etc.

Reading Lists