Oppositional concepts in population-based problem solving

点击翻译按钮获取中文摘要

Population-based optimization techniques such as genetic algorithms, differential evolution and ant colonies have diverse applications. These optimization methods have proven to be useful in many cases where conventional optimization methods encounter their applicability limits. However, population-based schemes have their own limitations. Specifically, they may often need considerable computational time to find a solution. This disadvantage becomes more visible the larger the population is, which is (almost) always the case when we deal with high-dimensional and/or complex optimization problems. Hence, methods by which to increase the speed of these techniques have been under investigation for quite some time. The main focus of this research will centre on development of methods to increase the speed of differential evolution and ant colonies. Oppositional concepts will be employed to accelerate the convergence of the methods while maintaining the necessary level of solution accuracy. Opposition-based approaches to optimization generally incorporate the simultaneous consideration of the solution and the opposite solution (chromosome and anti-chromosome, path and opposite path). Recent achievements in the successful design and use of opposition-based differential evolution encourage us to seek some fundamental answers with respect to a mathematical formalism for these techniques and to exploit the potentials of oppositional schemes for all population-based algorithms. Standard benchmark functions and metrics will be used to verify the better performance of opposition-based extensions of methods under investigation. As a real-world test case, segmentation of medical images, specifically breast and prostate ultrasound images, will be undertaken as well. Population-based methods have been used to extract objects from digital images in different ways. Their results, as reported in literature, are in some cases impressive. However, processing images with these methods are extremely expensive. This has restricted their use in practical cases. Any level of speedup is desirable here. Image data sets along with radiologist's ground-truth are available for experimental performance verification.