Development of Software System for State Machine Generation Using Genetic Algorithms

Abstract

Abstract—As a part of research works [1], [2], the developmentof a software system for state machine generation using geneticalgorithms [3] is being done. This software system increaseslevel of automation for designing state machine programs. Itwas named GAAP (Genetic Algorithms for Automata-basedProgramming). I. I NTRODUCTION For now many genetic algorithms realizations have beenoffered, and also genetic operators and chromosome repre-sentations [1]. Many of the existing genetic algorithms areonly theoretical evidence which is not supported by anysoftware solutions. Others ones are proved on test problemsbut hard to be applied for solving practical tasks due to lackof corresponding software systems flexibility.II. E XISTING SYSTEMS LIMITATIONS AND REQUIREMENTSFOR NEW ONES There are several common limitations of existing softwaresystems implementing genetic algorithms:1) Developed software system is strictly binded to a spe-cific problem at point of individual representation.2) Software implementation is done from scratch.3) Software system is restricted to modification and inte-gration with other software (particularly closed-source).4) Search result and intermediate population states cannotbe saved and therefore analyzed in future.As proved in NFL (No Free Lunch) theorem [4], allextremum search algorithms save the same efficiency averagedfor all possible fitness functions. Practical meaning of thistheorem is there is no silver bullet and that absolute success ofone optimization method in one field does not guarantee suchsuccess in other field. This indicates that for every specific fieldadditional research is needed to find most suitable optimizationmethod. All this implies following characteristics of softwaresystems based on genetic algorithms:1) Fitness function is built for each optimization task upona specific problem [1, sec. 2.3 and 2.4]. Hence softwaresystem should allow usage of different fitness function.