Citation Context

...cally improve an existing program using genetic programming. Typically, genetic improvement has focussed on changes using parts of the existing system. We develop the idea of software transplantation =-=[15]-=- and introduce the idea of GI as a means to specialise software. To investigate and experiment with GI for a particularly challenging problem, we selected the goal of using it to improve the execution...

Citation Context

... six times on recent nVidia GPU cards are reported compared to the original kernel on the same hardware. 1 Introduction Genetic Programming (GP) [1] is increasingly being used in Software Engineering =-=[2]-=-. We are using GP to make software more adaptable [3] and are particularly interested in GP to generate code [4; 5], for bug fixing [6] and for improving existing code [7; 8; 9; 10; 11; 12; 13]. With ...

Citation Context

... leave it to some later optimisation stage to decide on the exact parameters. However, Merrill et al. [9] were thinking of deterministic optimisation rather than more powerful search based techniques =-=[2]-=-. constantBasis directs the compiler to read pre-calculated spline coefficients from __constant__ memory rather than fetching them from the texture cache. It also forces the use of pre-calculated valu...

Citation Context

...ignificant speed gain was observed when the genetic pool was widened with multiple versions of the solver software [46]. 3.2.3 Migration and transplantation Code transplantation has been suggested in =-=[25]-=- for reverse engineering purposes such as inserting a feature from one system into another. It seems that GI has the tools and right approach to accomplish this task automatically by using GP and appr...

Citation Context

...promising for self-adaptive software systems, for several reasons. Recent work demonstrated the potential of SBSE techniques to solve problems at real-world scale, such as bug repair in large, extant software systems [27], [28]. This has resulted in a body of knowledge on how to scalably manipulate tree-based program constructs in terms of both representation and operator encodings (such as representing candidate program improvements as sets of changes rather than entire solutions [31]). Stochastic techniques have also recently been shown to successfully apply past knowledge to new situations [32], [33], suggesting promise in helping long-running systems evolve in response to new quality objectives or priorities. Stochastic searchbased techniques have been useful in developing unusual or unexpected solutions to difficult problems. The results of such techniques may therefore be informative for researchers trying to understand the design space of self-adaptive systems. Finally, although predicting and measuring hard-to-quantify quality attributes is an open problem, recent advances in probabilistic model checking (see, e.g., PRISM [34]) allows us to quantify expected costs and benefits ...

Citation Context

...sed to optimise learning and adaptivity. In particular, we see an important role for Genetic Programming (GP) [35] as a means of program improvement that has come to be known as ‘genetic improvement’ =-=[17, 18, 26]-=- (and has also been referred to as ‘evolutionary improvement’ [42]). 2. GENETIC IMPROVEMENT There has been a dramatic recent upsurge in interest and techniques for improving existing programs using ge...