JIANG De, QIU Yue, SHEN Hai-bin. Hardware design of kernel functions operation suitable for sensor systems[J]. 2013, 32(9): 93-96. DOI: 10.13873/j.1000-97872013.09.038.
适用于传感器系统的核函数运算硬件设计
摘要
针对人工智能与传感器技术结合的技术趋势
提出了2种适用于传感器系统的支持向量机核函数运算硬件实现方法。这2种方法均支持多种核函数运算
并采用了串行运算的方式
回避了原有架构对样本数不够灵活和面积对样本数过于敏感的缺点。对2种实现方法在多个角度进行对比
并探讨其优缺点。实验结果表明:一种适合于对面积要求高
支持核函数不多的场合;另一种适合于对速度要求高
支持核函数较多的场合。
Abstract
Aiming at trend of technology combining artificial intelligence and sensor
two kinds of hardware implementation methods of support vector machine(SVM) kernel fuction operation are presented.The two methods both support multiple operations of kernel functions and use serial operation style
which avoid the disadvantage of unflexibility to sample amount and excessive sensitivity of area to sample amount in previous structure.The two methods are compared in multiple perspectives
and advantages and disadvantages are discussed.Experimental result shows that one of them suits the situation of critical area limitation and supporting small amount of kernel functions
while the other one suits the situation of high-speed and supporting more amounts of kernel functions.