Abstract

Behavioural observations can often be described as a sequence of symbols drawn from a finite alphabet. However the inductive inference of such strings by any automated technique to produce models of the data is a non-trivial task. This paper considers modelling of behavioural data using probabilistic finite state automata (PFSAs). There are a number of information-theoretic techniques for evaluating possible hypotheses. The measure used in this paper is the Minimum Message Length (MML) of Wallace. Although attempts have been made to construct PFSA models by incremental addition of sub-strings using heuristic rules and the MML to give the lowest information cost, the resultant models cannot be shown to be globally optimal. Fogel's Evolutionary Programming can produce globally optimal PFSA models by evolving data structures of arbitrary complexity without the requirement to encode the PFSA into binary strings as in Genetic Algorithms. However, evaluation of PFSAs during the evolution pro...

Citations