A Genetic Algorithm for Locating Acceptable Structure Models of Systems (Reconstructability Analysis)
Cavallo, Roger; Advisor
Reale, Michael; Reviewer
Sengupta, Saumendra; Reviewer
computer and information science , general systems theory , computer and systems science , structure model , genetic algorithms , R (programming language)
The emergence of the field of General Systems Theory (GST) can be best attributed to the belief that all systems, irrespective of context, share simple, organizational principles capable of being mathematically modeled with any of many forms of abstraction. Structure modeling is a well‐developed aspect of GST specializing in analyzing the structure of a system ‐ that is, the interactions between the attributes of a system. These interactions, while intuitive in smaller systems, become increasingly difficult to comprehend as the number of measurable attributes of a system increases. To combat this, one may approach an overall system by analyzing its various subsystems and, potentially, reconstruct properties of that system using knowledge gained from considering a collection of these subsystems (a structure model). In situations where the overall system cannot be fully reconstructed based on a given structure model, the benefits and detriments associated with using such a model should both be considered. For example, while a model may be simpler to understand, or require less storage space in memory than the system as a whole, all information regarding that system may not be inferable from that model. As systems grow in size, determining the acceptability of every meaningful structure model of a system in order tofind the most acceptable becomes exceedingly resource-intensive. In this thesis, a measure of the memory requirements associated with storing a system or a set of subsystems (a structure model) is defined and is used in defining an objective measure of the acceptability of a structure as a representation of an overall system. A Genetic Algorithm for Locating Acceptable Structures (GALAS) is then outlined, with this acceptability criterion serving as an optimizable fitness function. The goal of this heuristic is to search the set of all meaningful structure models, without the need for exhaustively generating each, and produce those that are the most acceptable, based on predefined acceptability criteria.
Master of Science Thesis in Computer and Information Science, Department of Computer and Information Science, SUNY Polytechnic Institute. Approved and recommended for acceptance as a project in partial fulfillment of the requirements for the degree of Master of Science in Computer Science. Submitted for archiving in October 2018.