摘要
针对硬件进化 (evolvablehardware,EHW)存在的进化规模的扩展能力问题 ,研究了以内嵌式EHW的自适应要求为条件的染色体表示及计算复杂性问题 .为提高遗传机器学习计算效率 ,根据FPGA(fieldprogrammablegatearray)内部的结构特点 ,将可重构硬件的结构映射为遗传学习的染色体表示 ,提出一种符合EHW要求的二维染色体的遗传机器学习方法———ISPitts,构造了一种动态遗传机器学习框架 .实验结果显示 ,新方法不仅完成了四位比较器的内嵌式EHW实现 ,而且具有较高的进化效率 .
Aiming at the scalability problem of evolvable hardware, the chromosome representation method and computation complexity of embedded evolvable hardware on condition of self-adaptation was analyzed. To improve the efficiency of genetic machine learning, the architecture of a reconfigurable platform was mapped to the chromosome for genetic machine learning according to the structural characteristic of FPGA(field programmable gate array). Then a genetic learning method called ISPitts for a 2-dimensional expression of chromosome suitable for evolvable hardware was proposed, and a dynamic frame for genetic machine learning was presented. The proposed model can not only complete the 4-bit comparator in embedded style, but also exhibit much better efficiency than the functional abstract model in comparative experiments.
基金
国家自然科学基金资助项目 (699710 2 2)
关键词
硬件进化
遗传算法
机器学习
匹兹堡方法
Evolvable Hardware
Genetic Algorithm
Machine Learning
Pittsburgh-style Method
