Representation and Hidden Bias II Eliminating Defining Length Bias in Genetic Search via Sh.pdfVIP

Representatio n a n d Hidde n Bia s I I : E l i m i n a t i n g Definin g L e n g t h Bia s i n Geneti c Searc h vi a Shuffl e Crossove r Abstrac t The tw o most commonly employe d genetic search operator s are crossover an d mutation. Cross- Th e traditiona l crossove r operato r used i n genetic over produces offspring (new tria l solutions ) by search exhibit s a position-dependent bias called th e recombining th e informatio n fro m tw o parent s i n dcfining-length bias. We show how thi s bias results the manne r illustrate d i n Figur e 1 . It i s th e majo r in hidden biases tha t are difficul t t o anticipat e an d exploratory mechanis m of th e GA . Mutatio n compensate for . We introduc e a new crossove r prevents convergence of th e populatio n by flipping a operator , shuffle crossover, tha t eliminate s th e posi- smal l numbe r of randoml y selected bit s t o continu - tio n dependent bias of th e traditiona l crossove r ously introduc e variation . operato r by shuffling th e representatio n prio r t o applying crossover . W e also present experimenta l results tha t show tha t shuffle crossove r outperform s traditiona l crossove r o n a suite of five functio n optimizatio n problems . 1 . Introductio n Th e selected knowledg e representatio n serves as a