| 鲜开军*,丁新虎*,朱城超**,朱钟华**,徐东伟**.基于遗传算法的神经网络等价模型构建[J].高技术通讯(中文),2021,31(11):1136~1144 |
| 基于遗传算法的神经网络等价模型构建 |
| Neural network equivalent model construction based on genetic algorithm |
| |
| DOI：10.3772/j.issn.1002-0470.2021.11.003 |
| 中文关键词: 神经网络； 深度学习； 遗传算法（GA）； 等价模型 |
| 英文关键词: neural network, deep learning, genetic algorithm (GA), equivalent model |
| 基金项目: |
| 作者 | 单位 | | 鲜开军* | | | 丁新虎* | | | 朱城超** | | | 朱钟华** | | | 徐东伟** | |
|
| 摘要点击次数: 1485 |
| 全文下载次数: 1106 |
| 中文摘要: |
| 神经网络模型结构作为深度学习的重要组成部分，在很大程度上决定着深度学习的性能表现。而目前基于深度学习的应用，大部分都由经典的网络模型修改而来。由于无法获得原神经网络模型结构，本文根据原模型的输入输出数据以及经典的神经网络模型结构，构建了原模型的预测模型。该方法主要通过对预测模型的结构参数进行编码，并利用遗传算法（GA）进行选择、交叉、变异操作，从而构建出原模型的等价模型。对于同一输入数据，等价模型和原模型的输出基本保持一致。本文提出的构建方法在图像分类、信号调制类型分类和网络链路预测领域均取得了较好的效果。 |
| 英文摘要: |
| As an important part of deep learning, the structure of neural network model determines the performance of deep learning to a large extent. At present, most applications based on deep learning are modified from classic network models. Since the structure of original neural network model cannot be obtained, this paper constructs a prediction model of the original model based on the input and output data of the original model and the classic structure of neural network model. This method mainly encodes the structural parameters of the prediction model, and uses genetic algorithm (GA) to perform selection, crossover, and mutation operations to construct an equivalent model of the original model. For the same input data, the output of the equivalent model and the original model are basically the same. The construction method proposed in this paper has achieved good results in the fields of image classification, signal modulation type classification and network link prediction. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
