| 最小窥视孔长短时记忆模型 |
| |
| 引用本文： | 包志强,赵研,胡啸天,赵媛媛,黄琼丹.最小窥视孔长短时记忆模型[J].计算机工程与设计,2020,41(1):134-138. |
| |
| 作者姓名： | 包志强 赵研 胡啸天 赵媛媛 黄琼丹 |
| |
| 作者单位： | 西安邮电大学通信与信息工程学院,陕西西安710121;西安邮电大学通信与信息工程学院,陕西西安710121;西安邮电大学通信与信息工程学院,陕西西安710121;西安邮电大学通信与信息工程学院,陕西西安710121;西安邮电大学通信与信息工程学院,陕西西安710121 |
| |
| 基金项目： | 陕西省重点研发计划基金项目;陕西省教育厅专项科研项目 |
| |
| 摘 要： | 由于循环神经网络拥有复杂的模型结构,使训练模型达到最优变得困难。因此,提出一种最小窥视孔长短时记忆模型,它只有一个唯一门来更新信息,拥有两个网络层,通过减少一定的模型参数降低模型训练的难度,提高模型性能。实验结果表明,在不同数据集上,该模型性能高于长短期记忆模型,部分高于门循环单元模型,在参数个数、运行时间方面,其远小于长短期记忆模型以及门循环单元模型。
|
| 关 键 词： | 深度学习 循环神经网络 长短时记忆模型 门循环单元模型 最小窥视孔长短时记忆模型 |
Minimal peephole long short-term memory |
| |
| BAO Zhi-qiang,ZHAO Yan,HU Xiao-tian,ZHAO Yuan-yuan,HUANG Qiong-dan.Minimal peephole long short-term memory[J].Computer Engineering and Design,2020,41(1):134-138. |
| |
| Authors: | BAO Zhi-qiang ZHAO Yan HU Xiao-tian ZHAO Yuan-yuan HUANG Qiong-dan |
| |
| Affiliation: | (School of Communications and Information Engineering,Xi’an University of Posts and Telecommunications,Xi’an 710121,China) |
| |
| Abstract: | Finding the best practices for recurrent neural network(RNN)learning is a difficult task,because there are many complex model structures in RNN.A model was proposed,named as minimal peephole long short-term memory(MP-LSTM).It contained one unique gate to update data and two network layers.The design of MP-LSTM that decreased model parameters was useful to the training.Experimental results show that,in different datasets,the performance of the proposed model is better than that of long short-term memory(LSTM)model,and some of them are better than that of gated recurrent unit(GRU)model.The number of parameters and running time of the method are much smaller than that of the LSTM model and the GRU model. |
| |
| Keywords: | deep learning recurrent neural network(RNN) long short-term memory(LSTM) gated recurrent unit(GRU) minimal peephole long short-term memory(MP-LSTM) |
| 本文献已被 维普 万方数据 等数据库收录！ |
|
