欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
近年来图形处理器（GPU）快速拓展的可编程性能力加上渲染流水线的高速度及并行性，使得图形处理器通用计算（GPGPU）迅速成为一个研究热点。针对大规模神经网络BP算法效率低下问题，提出了一种GPU加速的神经网络BP算法。将BP网络的前向计算、反向学习转换为GPU纹理的渲染过程，从而利用GPU强大的浮点运算能力和高度并行的计算特性对BP算法进行求解。实验结果表明，在保证求解结果准确度不变的情况下，该方法运行效率有明显的提高。
英文摘要：
Recently, the raw speed, highly data-parallel nature and rapidly expanding programmability of graphics processing units make them an attractive platform for general purpose computation. As the efficiency of BP algorithm in large-scale neural network is relative low, proposed a GPU accelerated BP algorithm of neural network. Translated the forward computing and back-propagation learning of BP algorithm into texture rendering of GPU so as to solve the BP algorithm using the powerful float-point operation ability and high parallel computing characteristic of GPU. Experimental results show that the proposed method greatly raises the speed of BP algorithm in large-scale neural networks without losing the accuracy.
同期刊论文项目
同项目期刊论文
期刊信息
位置：