Rule-based training of neural networks

This paper describes RULEIN/RULEX; an automated technique for the refinement of a knowledge base. RULEIN constructs a Rapid Backprop (RBP) network from an initial, partially complete/accurate knowledge base which is formulated as a set of propositional rules. The RBP network is then trained on a set of examples drawn from the problem domain. RULEX is then applied to the weights of the trained network to extract a set of !refined propositional rules. The refined rule set represents the original knowledge base modified in the light of network training. Network training has the potential to remove inaccuracies in the original rule base, supplement partially correct initial rules, and add new rules. Rule initialisation can also speed up network learning by obviating the necessity of starting training from a tabula rasa configuration. RULEIN/RULEX is evaluated using rule !quality criteria and results are presented for some benchmark problems. The method has application in many areas but is particularly suited to overcoming the so called !knowledge acquisition bottleneck in the knowledge engineering phase of rule based expert system construction.

Decision Support Systems, 1996

This research explores a new approach to integrate neural networks and expert systems. The integrated system combines the strength of rule-based semantic structure and the learning capability of connectionist architecture. In addition, the approach allows users to define logical operators that behave much similar to that of human expert decision making process. Neural Logic Network (NEULONET) is used as the underlying building unit. A rule-based shell like environment is developed. The shell is used to built a prototype expert decision support system for future bonds trading. The system also provides a way to behave like different experts responding to different users and giving advice according: to different environmental situations.

Expert Systems with Applications, 2004

In this paper, we present an approach that integrates symbolic rules, neural networks and cases. To achieve it, we integrate a kind of hybrid rules, called neurules, with cases. Neurules integrate symbolic rules with the Adaline neural unit. In the integration, neurules are used to index cases representing their exceptions. In this way, the accuracy of the neurules is improved. On the other hand, due to neurule-based efficient inference mechanism, conclusions can be reached more efficiently. In addition, neurule-based inferences can be performed even if some of the inputs are unknown, in contrast to symbolic rule-based inferences. Furthermore, an existing symbolic rule-base with indexed exception cases can be converted into a neurule-base with corresponding indexed exception cases. Finally, empirical data can be used as a knowledge source, which facilitates knowledge acquisition. We also present a new high-level categorization of the approaches integrating rule-based and case-based reasoning. q

Applied Intelligence, 2000

As the second part of a special issue on “Neural Networks and Structured Knowledge,” the contributions collected here concentrate on the extraction of knowledge, particularly in the form of rules, from neural networks, and on applications relying on the representation and processing of structured knowledge by neural networks. The transformation of the low-level internal representation in a neural network into higher-level knowledge or information that can be interpreted more easily by humans and integrated with symbol-oriented mechanisms is the subject of the first group of papers. The second group of papers uses specific applications as starting point, and describes approaches based on neural networks for the knowledge representation required to solve crucial tasks in the respective application. The companion first part of the special issue [1] contains papers dealing with representation and reasoning issues on the basis of neural networks.

Knowledge Acquisition, 1991

In this paper, we argue that techniques proposed for combining empirical and explanation-based learning methods can also be used to detect errors in rule-based expert systems, to isolate the blame for these errors to a small number of rules and suggest revisions to the rules to eliminate these errors. We demonstrate that FOCL, an extension to Quinlan's FOIL program, can learn relational concepts in spite of an incorrect domain theory (e.g. a knowledge base of an expert system that contains some erroneous rules). A prototype knowledge acquisition tool, KR-FOCL, has been constructed that utilizes a trace of FOCL to suggest revisions to a rule base.

IFAC Proceedings Volumes, 1983

Knowledge acquisltlon is a crucial problem in the design of expert systems. In fact, bridging the gap between domain experts and expert system designers in constructing and mantaining a large rule base is a very demanding issue. In this paper we propose an approach to knowledge acquisition which is based on a deep rp.structuring of the usual architecture of an expert system. This includes: the partitioning of knowledge in a rule base and a strategy-rule base; the proposal of a flexible rule structure, the treatment of partial matching, the introduction of the new concept of goal in the inference process, and the design of a specific knowledge acquisition subsystem. The above issues are discussed in connection with a real application concerning safety in route transportation of hazardous materials.

IJCNN'99. International Joint Conference on Neural Networks. Proceedings (Cat. No.99CH36339)

Local basis function networks are a useful category of classifiers, with known variations developed in neural networks, machine learning and statistics communities. The localized range of activation of the hidden units have many similarities with rule-based representations. Neurofuzzy systems are a common example of an framework that explicitly integrates these approaches. Following this concept, we study alternatives for the development of rule-neural hybrid systems with the purpose of inducing robust and interpretable classifiers. Local fitting of parameters is done by a gradient descent optimization that modifies the covering produced by a rule induction algorithm. Two tasks are accomplished: how to select a small number of rules and how to improve precision. Accuracy is not the only target: indeed, the use of this architecture is better suited when one wants to achieve a good compromise between classification performance and simplicity.

Expert networks are networks of neural objects derived from expert systems. The hybrid nature of such networks allows the expert knowledge to be refined and augmented using sample data. The benefit of combining expert systems with neural network-like learning from data has been illustrated in a number of diagnosis and decision-making problem domains. The ability of these systems to learn illuminates knowledge previously unknown to human experts. This paper contains a synopsis of work in expert networks over the last several years, some of the findings wehave found useful and interesting, and an indication of current directions for this research.

Expert Systems, 2002

The aim of this research is to develop a model for designing rule-based expert systems that uses the forward chaining method of inference. The striking aspect of this model is that the inference engine is based on a simple representation of rules and facts in relational database tables. Rules are decomposed and represented in tables at two levels, which allow the developing of expert systems in any programming language that supports SQL. The explanation facility uses tables containing the explanations of the result of each rule. The model proposed in this paper is based on a simple approach to represent facts and rules in relational database tables. The advantage of this model lies in focusing the design of rule-based expert systems toward knowledge representation in a relational database, reducing effort and programming difficulties.