摘要:

Abstract:A multilayer perceptron is known to be capable of approximating any smooth function to any desired accuracy if it has a sufficient number of hidden neurons. But its training, based on the gradient method, is usually a time consuming procedure that may converge toward a local minimum, and furthermore its performance is greatly influenced by the number of hidden neurons and their initial weights. Usually these crucial parameters are determined based on the trial and error procedure, requiring much experience on the designer's part.In this paper, a constructive design method (CDM) has been proposed for a two‐layer perceptron that can approximate a class of smooth functions whose feature vector classes are linearly separable. Based on the analysis of a given data set sampled from the target function, feature vectors that can characterize the function‘well’are extracted and used to determine the number of hidden neurons and the initial weights of the network. But when the classes of the feature vectors are not linearly separable, the network may not be trained easily, mainly due to the interference among the hyperplanes generated by hidden neurons. Next, to compensate for this interference, a refined version of the modular neural network (MNN) has been proposed where each network module is created by CDM. After the input space has been partitioned into many local regions, a two‐layer perceptron constructed by CDM is assigned to each local region. By doing this, the feature vector classes are more likely to become linearly separable in each local region and as a result, the function may be approximated with greatly improved accuracy by MNN. An example simulation illustrates the improvements in learning speed using a smaller number of

ISSN:0266-4720    

DOI:10.1111/j.1468-0394.1996.tb00118.x

出版商:Blackwell Publishing Ltd    

年代:1996

数据来源: WILEY

摘要:

Abstract:This paper proposes artificial neural networks (ANN) as a tool for nonlinear combination of forecasts. In this study, three forecasting models are used for individual forecasts, and then two linear combining methods are used to compare with the ANN combining method. The comparative experiment using real‐world data shows that the prediction by the ANN method outperforms those by linear combining methods. The paper suggests that the ANN method can be used as an alternative to conventional linear combining methods to achieve greater forecasting accurac

ISSN:0266-4720    

DOI:10.1111/j.1468-0394.1996.tb00119.x

出版商:Blackwell Publishing Ltd    

年代:1996

数据来源: WILEY

摘要:

Abstract:This paper presents work on an interactive fault diagnosis expert system for a helpdesk application. A knowledge representation and inference algorithm is proposed to satisfy three design specifications: (1) no parallel event exists in human fault reporting; (2) the diagnostic sequence is unpredictable, and (3) the inference engine is passive in an event‐driven environment. A lattice data structure is designed for knowledge representation, which is generated automatically from a script of decision rules. The inference engine works in a transaction‐like style by prompting and responding to the user according to the knowledge in the lattice. It can explicitly guide the inference sequence, as well as respond to ad hoc input from the u

ISSN:0266-4720    

DOI:10.1111/j.1468-0394.1996.tb00120.x

出版商:Blackwell Publishing Ltd    

年代:1996

数据来源: WILEY

摘要:

Abstract:A knowledge base management system (KBMS) realises a combination of techniques found in database management systems and knowledge‐based systems. At the data model and knowledge representation level, many systems of this kind constitute a marriage of the relational data model and the rule‐based reasoning. Experience has shown that either approach is restricted in the way it can express the demanding information and knowledge structures required for applications like decision support systems. Two new technologies offer an exciting new integrated approach to knowledge management. Object‐oriented database management systems (OODBMS) provide an object model that supports powerful abstraction mechanisms to facilitate the modelling of highly structured information. Whereas case‐based reasoning (CBR) systems are knowledge bases which organise their capabilities around a memory of past cases and the notion of similarity. Both types of system are built upon two fundamental concepts: 1) the retrieval of entities with potentially complex structure, called objects in the former, and cases in the latter type of system; 2) the organisation of those entities in collections with common characteristics. In an OODBMS such collections are termed extents, and in CBR they are usually called categories. In either system, the conceptual meta notion to represent both, objects as well as extents, and cases as well as categories, is the class.Revolving around a Conceptual Case Class and extending a standard object model, this paper proposes a novel and general approach to represent case‐knowledge and to build KBMSs. The work presented here is a spin‐off of the design of an object query language within the ESPRIT p

ISSN:0266-4720    

DOI:10.1111/j.1468-0394.1996.tb00121.x

出版商:Blackwell Publishing Ltd    

年代:1996

数据来源: WILEY

摘要:

The Handbook of Brain Theory and Neural NetworksMichael A. Arbib (Ed.)Virtual Reality SystemsJohn VinceThe Uncertain Reasoner's CompanionJ. B. Paris

ISSN:0266-4720    

DOI:10.1111/j.1468-0394.1996.tb00122.x

出版商:Blackwell Publishing Ltd    

年代:1996

数据来源: WILEY

ISSN:0266-4720    

DOI:10.1111/j.1468-0394.1996.tb00123.x

出版商:Blackwell Publishing Ltd    

年代:1996

数据来源: WILEY