No description available.

No description available.

No description available.

No description available.

Many expert, or knowledge-based, systems have been constructed in the domain of ecology, several of which are concerned with habitat evaluation. However, these systems have been geared to solving particular problems, with little regard paid to the underlying relationships that exist within a biological system. The implementation of problem-solving methods with little regard to understanding the more primary knowledge of a problem area is referred to in the literature as 'shallow', whilst the representation and utilisation of knowledge of a more fundamental kind is termed 'deep'. This thesis contains the details of a body of research exploring issues that arise from the refinement of traditional expert systems methodologies and theory via the incorporation of depth, along with enhancements in the sophistication of the methods of reasoning (and subsequent effects on the mechanisms of communication between human and computer), and the handling of uncertainty. The approach used to address this research incorporates two distinct aspects. Firstly, the literature of 'depth', expert systems in ecology, uncertainty, and control of reasoning and related user interface issues are critically reviewed, and where inadequacies exist, proposals for improvements are made. Secondly, practical work has taken place involving the construction of two knowledge based systems, one 'traditional', and the other a second generation system. Both systems are primarily geared to the problem of evaluating a pond site with respect to its suitability for the great crested newt (Triturus cristatus). This research indicates that it is possible to build a second-generation knowledge-based system in the domain of ecology, and that construction of the second generation system required a magnitude of effort similar to the firstgeneration system. In addition, it shows that, despite using different architectures and reasoning strategies, such systems may be judged as equally acceptable by endusers, and of similar accuracy in their conclusions. The research also offers guidance concerning the organisation and utilisation of deep knowledge within an expert systems framework, in both ecology and in other domains that have a similar concept-rich nature.

Knowledge Acquisition has long been acknowledged as the bottleneck in producing Expert Systems. This is because, until relatively recently, the KA (Knowledge Acquisition) process has concentrated on extracting knowledge from a domain expert, which is a very time consuming process. Support tools have been constructed to help this process, but these have not been able to reduce the time radically. However, in many domains, the expert is not the only source of knowledge, nor indeed the best source of knowledge. This is particularly true in industrial settings where performance information is routinely archived. This information, if processed correctly, can provide a substantial part of the knowledge required to build a KB (Knowledge Base). In this thesis I discuss current KA approaches and then go on to outline a methodology which uses KD (Knowledge Discovery) techniques to mine archived time series data to produce fault detection and diagnosis KBs with <I>minimal expert input. </I>This methodology is implemented in the TIGON system, which is the focus of this thesis. TIGON uses archived information (in TIGON's case the information is from a gas turbine engine) along with <I>guidance</I> from the expert to produce KBs for detecting and diagnosing faults in a gas turbine engine. TIGON's performance is also analysed in some detail. A comparison with other related work is also included.

by Wong Man Leung. / Thesis (M.Phil.)--Chinese University of Hong Kong, 1990. / Bibliography: leaves 200-206. / ABSTRACT --- p.A-1 / Chapter 1 --- Introduction --- p.1 / Chapter 1.1 --- Overview --- p.1 / Chapter 1.2 --- Properties of an ideal AKARS --- p.4 / Chapter 1.3 --- The architecture of AKARS-1 --- p.9 / Chapter 1.4 --- Research approach --- p.13 / Chapter 2 --- state of the art of machine learning --- p.16 / Chapter 2.1 --- Learning by instruction --- p.16 / Chapter 2.2 --- Learning by analogy --- p.17 / Chapter 2.3 --- Learning from examples --- p.17 / Chapter 2.4 --- Learning from observation and discovery --- p.24 / Chapter 3 --- SESS: a Simple Expert System Shell --- p.26 / Chapter 3.1 --- Overview of SESS --- p.26 / Chapter 3.2 --- Knowledge representation --- p.27 / Chapter 3.2.1 --- Representation of attribute --- p.27 / Chapter 3.2.2 --- Representation of fuzzy concepts --- p.28 / Chapter 3.2.3 --- Representation of rules --- p.29 / Chapter 3.3 --- Reasoning in SESS --- p.31 / Chapter 3.3.1 --- Rule evaluation --- p.31 / Chapter 3.3.2 --- Rules with multiple antecedent conditions --- p.32 / Chapter 3.3.3 --- Calculation of certainty factor --- p.33 / Chapter 4 --- A prototypical learning component --- p.35 / Chapter 4.1 --- A prototypical Automatic Knowledge Acquisition System: AKA-1 --- p.35 / Chapter 4.1.1 --- Introduction to AKA-1 --- p.35 / Chapter 4.1.2 --- A generic rule learning algorithm --- p.37 / Chapter 4.1.3 --- Method for evaluating the discriminatory abilities --- p.43 / Chapter 4.1.4 --- Method for determining the best attribute/value pair --- p.44 / Chapter 4.1.4.1 --- Determining the best nominal attribute/value pair --- p.44 / Chapter 4.1.4.2 --- Determining the best structural attribute/value pair --- p.46 / Chapter 4.1.4.3 --- Determining the best linear attribute/value pair --- p.48 / Chapter 4.1.5 --- Method for calculating certainty factors of rules --- p.50 / Chapter 4.1.6 --- Rule inducing algorithm of AKA-1 --- p.50 / Chapter 4.2 --- Generalizing production rules --- p.51 / Chapter 4.2.1 --- Testing significance of condition --- p.52 / Chapter 4.2.1.1 --- Chi-squre test (Large sample test) --- p.53 / Chapter 4.2.1.2 --- Fisher-Irwin's exact test (Small sample test) --- p.54 / Chapter 4.2.2 --- The generalization algorithm --- p.55 / Chapter 4.3 --- Case studies --- p.55 / Chapter 4.3.1 --- Case one --- p.56 / Chapter 4.3.2 --- Case two --- p.59 / Chapter 4.3.3 --- Case three --- p.59 / Chapter 4.3.4 --- Comparison with ID3 --- p.60 / Chapter 5 --- inducing fuzzy rules from inexact examples --- p.62 / Chapter 5.1 --- Introduction to AKA-2 --- p.62 / Chapter 5.2 --- Notations --- p.63 / Chapter 5.3 --- Method for selecting attribute/value pairs in AKA-2 --- p.67 / Chapter 5.4 --- Evaluating certainty factors of rules --- p.72 / Chapter 6 --- HERES: a HEuristic REfinement System --- p.78 / Chapter 6.1 --- Introduction to HERES --- p.78 / Chapter 6.2 --- Refinement concepts of HERES --- p.79 / Chapter 6.2.1 --- Refinement operations --- p.80 / Chapter 6.2.2 --- Refinement phases of HERES --- p.81 / Chapter 6.2.3 --- Strategy for knowledge base refinement --- p.83 / Chapter 6.2.4 --- Refinement examples --- p.86 / Chapter 6.2.5 --- Performance statistics --- p.91 / Chapter 6.2.6 --- Rule statistics --- p.94 / Chapter 6.2.7 --- Summary of refinement concepts --- p.96 / Chapter 6.3 --- Logical structure of HERES --- p.98 / Chapter 6.4 --- Rule analysis --- p.101 / Chapter 6.4.1 --- Rule analysis for generalization --- p.101 / Chapter 6.4.1.1 --- For rules with non-fuzzy final conclusions --- p.101 / Chapter 6.4.1.2 --- For hierarchical rules with non-fuzzy final conclusion --- p.107 / Chapter 6.4.1.3 --- For rules with fuzzy final conclusions --- p.108 / Chapter 6.4.2 --- Rule analysis for specialization --- p.108 / Chapter 6.4.2.1 --- For rules with non-fuzzy final conclusions --- p.108 / Chapter 6.4.2.2 --- For rules with fuzzy conclusions --- p.111 / Chapter 6.5 --- Modification of rule statistics --- p.111 / Chapter 6.6 --- First order G-3 refinement --- p.117 / Chapter 6.7 --- Higher order G-S refinement --- p.120 / Chapter 6.8 --- Heuristics of HERES --- p.122 / Chapter 6.8.1 --- Notations --- p.123 / Chapter 6.8.2 --- Control heuristics --- p.124 / Chapter 6.8.3 --- Strategic heuristics --- p.126 / Chapter 6.8.3.1 --- Strategic heuristics for generalization --- p.127 / Chapter 6.8.3.2 --- Strategic heuristics for specialization --- p.127 / Chapter 6.8.4 --- Refinement heuristics --- p.128 / Chapter 6.8.4.1 --- Refinement heuristics for generalization --- p.128 / Chapter 6.8.4.2 --- Refinement heuristics for specialization --- p.131 / Chapter 6.9 --- Discussion --- p.132 / Chapter 7 --- Verification of AKARS-l --- p.134 / Chapter 7.1 --- Verification methodologies --- p.134 / Chapter 7.1.1 --- Existing examples methodology --- p.134 / Chapter 7.1.2 --- Existing knowledge base methodology --- p.135 / Chapter 7.2 --- Methods for evaluating the performance of a knowledge base --- p.141 / Chapter 7.2.1 --- First method (Successful rate method) --- p.141 / Chapter 7.2.2 --- Second method (MDCF method) --- p.145 / Chapter 7.3 --- Case studies --- p.149 / Chapter 7.3.1 --- Case one --- p.149 / Chapter 7.3.2 --- Case two --- p.152 / Chapter 7.3.3 --- Case three --- p.156 / Chapter 7.4 --- Verification results --- p.157 / Chapter 7.4.1 --- Results of case one --- p.157 / Chapter 7.4.1.1 --- First experiment --- p.157 / Chapter 7.4.1.2 --- Second experiment --- p.158 / Chapter 7.4.1.3 --- Discussion --- p.160 / Chapter 7.4.2 --- Results of case two --- p.162 / Chapter 7.4.2.1 --- First experiment --- p.162 / Chapter 7.4.2.2 --- Second experiment --- p.163 / Chapter 7.4.2.3 --- Third experiment --- p.165 / Chapter 7.4.2.4 --- Fourth experiment --- p.166 / Chapter 7.4.2.5 --- Discussion --- p.167 / Chapter 7.4.3 --- Results of case three --- p.169 / Chapter 7.5 --- Discussion --- p.172 / Chapter 8 --- Developing hierarchical knowledge bases --- p.173 / Chapter 8.1 --- Introduction --- p.173 / Chapter 8.2 --- Acquire hierarchical knowledge bases by using AKARS-1 --- p.176 / Chapter 8.2.1 --- Difficulties of inducing multiple-level rules --- p.176 / Chapter 8.2.2 --- Structural approach of building hierarchical knowledge bases --- p.179 / Chapter 8.3 --- Further improvement on AKARS-1 to induce hierarchical knowledge bases --- p.181 / Chapter 9 --- Conclusion --- p.182 / Appendix A --- p.186 / Appendix B --- p.187 / Appendix C --- p.188 / Appendix D --- p.193 / Appendix E --- p.194 / Reference --- p.200

by Wong Man Hon. / Thesis (M.Phil.)--Chinese University of Hong Kong, 1989. / Bibliography: leaves 197-202.

by Shu-kin Kwan. / Thesis (M.Ph.)--Chinese University of Hong Kong, 1988. / Bibliography: leaves 151-156.

Individuals are increasingly using expert system tax programs as a substitute for paid professionals when preparing their income tax returns. This study examines ways that expert systems encourage the same aggressive results documented when paid professionals are used. Examining the use of expert systems and the related behavior of taxpayers reveals aggressive reporting related to the commonly used warning alerts in tax programs. Using an experimental economics setting in which participants report liabilities with the possibility of penalties for noncompliant reporting, participants filled out a Claim Form mimicking a Schedule C in one of four conditions: manual preparation, no alerts, alerts triggered at a high threshold of reporting aggression, and alerts triggered at a low level of reporting aggression. Comparing the amounts deducted in each condition revealed that warning alerts with low thresholds of activation decreased aggressive reporting while warning alerts with high thresholds of activation increased aggressive reporting. Survey instruments measuring user satisfaction indicated significantly lower satisfaction when (high or low level) warning alerts were used versus no warning alerts. Contrary to expectations, respondents using the expert system tax program with high threshold warning alerts compared to no warning alerts reported a significantly higher perception of accuracy. This study demonstrates the extreme to which taxpayers are swayed by perceived aspects of the tax software that are irrelevant to the facts of their tax situations. Exactly what taxpayers need to be given by way of guidance and direction to comport their behavior to the tax laws is a critical question of public policy.