Verification and validation of knowledge-based systems

Abstract

Knowledge-based systems (KBS) are being used in many applications areas where their failures can be costly because of the losses in services, property, or even life. To ensure their reliability and dependability, it is therefore important that these systems are verified and validated before they are deployed. This paper provides perspectives on issues and problems that impact the verification and validation (V&V) of KBS. Some of the reasons V&V of KBS is difficult are presented. The paper also provides an overview of different techniques and tools that have been developed for performing V&V activities. Finally, some of the research issues that are relevant for future work in this field are discussed.

Towards a Software System Providing Knowledge about Energy and Resource Efficiency Potentials within the Product and Process Development of Electric Drives

“…The second step of the progressive knowledge transfer, the validation, is often neglected or even ignored when it comes to building a knowledge base [49,50]. This constitutes a serious mistake as the validation of a KBS has a major effect on the quality of the whole system [49,50,51]. The validation can function as a preventive measure and should be integrated into an early stage of the acquisition process [40,49].…”

An Enhanced Petri Net Model to Verify and Validate a Neural-Symbolic Hybrid System

“…The main goal of verification is to obtain the consistent, complete and correct system. Because of that, anomalies on KB must be detected (Ramaswamy, Sarkar, & Chen, 1997;Ramirez & De Antonio, 2001;Tsai, Vishnuvajjala, & Zhang, 1999;Yang, Tsai, & Chen, 2003). An anomaly is referred to common fault patterns according to an analysis technique (Ramirez & De Antonio, 2007;Tsai, Vishnuvajjala, & Zhang, 1999).…”

“…Because of that, anomalies on KB must be detected (Ramaswamy, Sarkar, & Chen, 1997;Ramirez & De Antonio, 2001;Tsai, Vishnuvajjala, & Zhang, 1999;Yang, Tsai, & Chen, 2003). An anomaly is referred to common fault patterns according to an analysis technique (Ramirez & De Antonio, 2007;Tsai, Vishnuvajjala, & Zhang, 1999). This KB can contain errors due to: 1) the existence of several human experts providing their experiences in the application field 2) the inserted knowledge can not be represented properly because of communication problems between human experts and the knowledge engineer, 3) the information may be missed during knowledge insertion due to matching of same one to the neural network, 4) The base of examples might be redundant due to a bad selection of samples, and 5) information may be missed or gained during the integration process of numerical and symbolic knowledge (Cruz, Reyes, Vergara, & Pinto, 2006;He, Chu, & Yang, 2003;Santos, 1998;Villanueva, Cruz, Reyes, & Benítez, 2006).…”

“…The validation process allows analyzing the quality of the KB and the possibility of obtaining right solutions to the problems of the domain (Knauf, Gonzalez, & Abel, 2002;Tsai, Vishnuvajjala, & Zhang, 1999).…”

“…Traditional Petri nets have inherent disadvantages. Some of these disadvantages are: deficiency of flexible descriptions for negative relations and necessity to formalize the original KB before beginning verification process, for example a logical system or production systems, due to the difficulty of expressing logical disjunctions (He, Chu, & Yang, 2003;Nazareth, 1993;Wu & Lee, 1997;Tsai, Vishnuvajjala, & Zhang, 1999;Yang, Lee, Chu, & Yang, 1998). In order to overcome these problems an enhanced Petri net model is proposed by us.…”