A Proposal for Including Behavior in the Process of Object Similarity Assessment with Examples from Artificial Life

Abstract

The similarity assessment process often involves measuring the similarity of objects X and Y in terms of the similarity of corresponding constituents of X and Y, possibly in a recursive manner. This approach is not useful when the verbatim value of the data is of less interest than what they can potentially “do,” or where the objects of interest have incomparable representations. We consider the possibility that objects can have behavior independent of their representation, and so two objects can look similar, but behave differently, or look quite different and behave the same. This is of practical use in fields such as Artificial Life and Automatic Code Generation, where behavior is considered the ultimate determining factor. It is also useful when comparing objects that are represented in different forms and are not directly comparable. We propose to map behavior into data values as a preprocessing step to Rough Set methods. These data values are then treated as normal attributes in the similarity assessment process.