The evolution of complete software systems

摘要: This thesis tackles a series of problems related to the evolution complete software systems both in terms underlying Genetic Programming system and application that system. A new representation is presented addresses some issues with other Program representations while keeping their advantages. This combines easy reproduction linear with inheritable characteristics tree by using fixed-length blocks of genes representing single program statements. means each block will always map same statement parent child unless it is mutated, irrespective changes surrounding blocks. method is compared variable length gene used other representations with clear improvement similarity between child. Traditionally, fitness functions have either been created as selection of sample inputs known outputs or hand-crafted evaluation functions. A new method creating introduced takes the formal specification desired function its basis. approach ensures complete concise. The functions created from specifications are simple input/output pairs results show specifications perform significantly better. A set list manipulation was evolved an application components. These have the common feature they all need be 100% correct useful. Traditional Programming mainly optimization or approximation problems. good but do highlight the problem scalability more complex lead dramatic increase required time. Finally, graphical user interfaces addressed. representation for based on for programs. In this case represents component user interface. interface determined comparing it series of constraints, which specify layout, style functionality requirements. A web-based desktop-based were evolved. With these approaches Programming, of complete software now realistic goal.