基于反向随机投影的神经网络改进算法

An improved neural network algorithm based on reverse random projection

提出基于反向随机局部投影的神经网络效率改进算法,通过降低深度学习中的网络规模,重点解决了从“局部连接”到“全连接”和随机节点抽取时输入端节点信息丢失的问题,从而提升网络的效率。在算法中设置缩减参数,提升了算法的可伸展性,以适用于不同数据集的学习。通过数据集ISOLET进行实验,结果表明,基于反向随机局部投影的神经网络效率改进算法的准确率、效率分别平均提升了3.48%和105.21%;在迭代20次的实验中进行了缩减参数调节实验,当参数设置为1.4时其准确率则优于传统全连接网络2.61%,效率提升了272.78%,具有明显的优势。

A neural network efficiency improvement algorithm based on reverse random local projection was proposed.By reducing the network size in deep learning,the problem of information loss of input nodes caused by“local connection”to“full connection”and random node extraction was solved,thus improving the efficiency of the network.The reduction parameters were set in the algorithm to improve the extensibility of the algorithm,so that it can be applied to the learning of different data sets.Then,experiments were conducted,using data set ISOLET.Results indicated that the accuracy and efficiency of the algorithm based on reverse random local projection were increased by 3.48%and 105.21%respectively.When echos were set as 20 the reduction parameter adjustment experiment was carried out,and when the parameter was set as 1.4,the accuracy was improved by 2.61%compared with the traditional fully-connected network,and the efficiency was improved by 272.78%,which showed obvious advantages.