Abduction can be considered as a principal way of reasoning for problem solving. Abductive inference consists in generation of hypotheses which explain — or logically imply — the phenomenon under investigation in view of accessible background knowledge and are consistent with all other observations. Looking for such hypotheses is typically performed with a spectrum of trial-and-error or search methods and tools. In case of purely logical statements the hypotheses take the form of a set of facts, both positive and negative ones. For example, in case of model based diagnostic reasoning, such diagnostic hypotheses can be generated by consistency based reasoning with minimal search effort. In more complex cases, where values of certain variables are to be found, pure backtracking search becomes inefficient. In this paper we attempt to put forward such abductive inference into a formal framework of Constraint Programming in order to enable the use of constraint propagation techniques. The main idea behind this approach is to make abduction more constructive. The discussion is illustrated with a diagnostic example of a multiplier-adder system.

Rule-based systems constitute the most popular tool for specification of operational knowledge in the majority of knowledge-based systems. This paper addresses the issue of analysis and verification of selected properties of such systems in a systematic way. A uniform, tabular form of single-level rule-based systems is put forward. Such systems can be used independently as a generalized form of decision tables, or as the lower level components of a hierarchical, multilevel control and decision support knowledge-based system. An algebraic knowledge representation form is proposed and algebraic bases for system verification are outlined. © 2001 John Wiley & Sons, Inc.

One of the key issues for efficient application of Case-Based Reasoning (CBR) methodology is to define an appropriate, powerful knowledge representation structure for case encoding and definition of flexible matching algorithms. The basic paradigm consisting in straightforward application of Relational Databases, although conceptually simple and computationally efficient, suffers from inflexible matching algorithms and the lack of possibilities to represent more complex data structures. This paper provides an analysis of certain possible extensions of data structures to be represented within the widely accepted relational database like paradigm, here called Tabular Systems , while providing new possibilities to represent quite complex patterns. As the basic underlying structure the concept of an object is used. A review of selected non-atomic data items, i.e. structures such as sets, record-like structures, lists, intervals, terms, trees, graphs etc. is provided. Flexibility of this structures comes from the fact that they can cover a number of particular cases. For such structures flexible partial matching procedures are proposed. The flexibility is achieved through admission of several types (levels) of matching, different with respect to qualitative perception of the level of similarity. Such flexible matching of data structures provides a way for generation of a match both with respect to parameter adjustment and structure manipulation.