摘要： 利用非专业图形卡来实现批量式自组织映射(batch version of the SOM)的学习过程。算法基于OpenGL的扩充规范NV_fragment_program,并使用了最新的浮点缓冲区扩充规范NV_float_buffer,使学习过程达到了与CPU运算一致的精度,并且提高了训练速度,缩短了学习时间。

关键词: Batch SOM, GPU计算, 图形硬件, 机器学习

Abstract: This paper describes the realization of learning process of batch SOMs based on commonly available graphics hardware accelerators. The algorithm is based on the extensions of OpenGL, the fragment shader NV_fragment_program. It also adopts the floating point buffer extension (NV_float_buffer) which makes the learning process the same precision as CPU does. Compared with CPU, the learning speed is improved and the learning time is reduced.

中图分类号: 

• TP301