Clifford Grossner

Validating dynamic properties of rule-based systems

Abstract Rule-based systems can be viewed as possessing two sets of properties: static and dynamic. Static properties are those that can be evaluated without executing the system, and dynamic properties can be evaluated only by examining how the system operates at run time. The dynamic properties of a rule-based system have been largely neglected in validation and verification work done thus far. Structural verification and static testing techniques do not yield information on how a rule-based system achieves its goals at run-time, given a set of input data. This paper presents a model for the relationship between the goal states achieved by a rule-based system, the set of inter-related rules that must fire to achieve each goal state, and the data items required for the rules in the rule sequence to fire. Then, we describe a method for applying this model to study the dynamic properties of a rule-based system. It is demonstrated that this model permits the validation of dynamic properties of a rule-based system, enabling system developers to decide: (1) if the manner in which the system pursues goals is valid according to the specifications (and expectations) of the designers; (2) what relationship exists between the quality of system output for a given test case and the goals achieved during problem-solving on that test case; and (3) how the overall problem-solving activity of the system relates to the availability of input data.

Structure-based validation of rule-based systems

A central problem in validating rule-based systems is ensuring that the structural components of the system have been tested sufficiently thoroughly. In this paper, we present a method that can be used for structure-based testing of rule-based systems. Our method includes a model for determining the structural components in a rule base, a metric for quantifying coverage (the degree to which the structural components in a rule base have been tested) and techniques for analyzing the coverage obtained when a rule base is exercised on a given test set. We present the results produced by applying our method for structure-based testing to a complex rule-based system that had been previously subjected to functional testing. The results we obtained indicate that our method for structure-based testing reveals several types of faults not shown by previous testing, and quantifies the extent to which the rule base has been tested.

An Integrated Workstation for the Visually Handicapped

Because this workstations device drivers are integrated into its resident CP/M operating system, it can run any CP/M-compatible program and produce output in synthesized speech or braille.

Incremental validation of policy-based systems

Management and control of systems are becoming more and more automated and based on policies that are expressed in a suitable policy specification language. Such policies need to be validated to ensure correct operation of the managed or controlled systems. The set of policies in a system is dynamic and is constantly changing. When this set is very large, incremental policy validation is a topic worth exploring. In this paper, such a notion of incremental policy-validation is introduced. The well-known decision tables, a systems analysis tool, are adapted for representing policies. Based on such adaptations, two algorithms are developed for incremental validation and they are compared, with respect to the number of computations they require.

CONSENSUS AND COMPROMISE: PLANNING IN COOPERATING AGENTS

Resolving disparate viewpoints via planning is an important aspect of the distributed problem solving performed by the agents in Cooperative Intelligent Information Systems. We propose a distributed planning protocol called Consensus. Consensus specifies a methodology by which agents exchange information indicating their preferred actions, integrate these different sets of actions, resolve any conflicts that exist, and choose a joint set of actions. Within the framework of the Consensus protocol, two different heuristics are proposed for resolving conflicts which overcome the computational complexity of an exhaustive search. An implementation of the Consensus protocol is analyzed experimentally to assess its performance in resolving conflicts. The experimental results indicate the trade-off between the cost of planning and the quality of the plan produced with respect to the heuristics used for resolving conflicts.

Cuenet - A Distributed Computing Facility

This flexible system, supporting any interconnection topology specified by a programmer, allows system analysts to experiment with alternate program decomposition strategies.

Design of an interactive data retrieval system for casual users

Abstract Among the important criteria in the design and implementation of an interactive data retrieval system for casual users are: a friendly user interface, aids to train beginners on the system, versatility in output media, and error recovery techniques in an interactive session. These factors are considered in the design of an interactive data retrieval system which is used for the study of commodity market data. Although the concepts and considerations discussed in this article are discipline specific, the methodologies adopted are general in nature. The proposed system has been implemented and tested using an experimental data base on gold and silver futures.

An Intelligent Braille Display Device

The high cost of braille output devices has prevented their wide use among the visually handicapped. Here, a much cheaper device is implemented.

Towards a tool for the design of cooperating expert systems

A tool to aid the designer of cooperative distributed problem solving (CDPS) systems will determine whether a proposed CDPS system violates any of the constraints for CDPS systems, and will display a prediction of the performance that can be expected from that CDPS system. This tool will rely on a predictive model for the performance of the expert systems in the CDPS system. Using the Blackbox Expert (a DAI testbed), an experiment was conducted to test the performance of expert systems in a CDPS system. Statistically significant differences in the Blackbox Experts performance were found when the CDPS system was changed. The results of the experiment are being using to create a predictive model for the performance of the Blackbox Expert when it is a member of the CDPS system.<<ETX>>