SOME RECENT RESULTS OF NON-DETERMINISTIC MODELLING OF BEHAVIOUR SEQUENCES

Abstract

This paper surveys several non-deterministic modelling techniques and places them in a uniform framework. Three basically different methods are discussed: enumeration and evaluation of possible models; reduction of a large model by coalescing states; and limited-context methods which analyse and process all strings of a given length which occur in the behaviour. Three results of recent work are presented. Firstly, the methodology of enumeration and evaluation is extended to the case where the model space permits recursion. Secondly, some new experiments on successive reduction of large models are described, which indicate that the quality of models produced by this technique is rather variable. Thirdly, it is shown that limited context methods which discard information about transitions cannot identify certain kinds of inputs. However, they can take advantage of assistance from a "teacher" in the form of markers in the input string.