摘要
以"中国视云"科研平台为依托,针对神经网络模型可视化展示,提出一种卷积神经网络(CNN)核函数可视化方法。该方法中,通过使用最大激活函数对神经网络核函数进行可视化计算并形成功能模块。实验结果表明:该方法能够清晰展示CNN核函数和资源占用变化,具有方便操作、泛用性高等特点。该方法可对CNN模型的解释和模型结构与参数改进提供参考和借鉴。
Based on the China Visual Cloud platform of Shanghai University,this paper proposes a new method of kernel function visualization with CNN for the visualization demonstration of neural network models.In this method,the kernel function of CNN is visualized by the maximum activation function and then formed the functional modules.The experimental results show that the proposed method can clearly demonstrate the changes of kernel function and resource occupancy in the CNN model,and has the characteristics of convenient operation and high generalization.It can provide reference for the interpretation of the CNN model and the improvement of the model structure and parameters.
作者
李成范
胡子荣
刘岚
丁雪海
童维勤
LI Chengfan;HU Zirong;LIU Lan;DING Xuehai;TONG Weiqin(School of Computer Engineering and Science,Shanghai University,Shanghai 200444,China;School of Electronic and Electrical Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)
出处
《实验室研究与探索》
CAS
北大核心
2021年第5期57-61,共5页
Research and Exploration In Laboratory
基金
上海市科委项目(19142201600,19dz2252600)。
关键词
中国视云
卷积神经网络
模块化
可视化
核函数
China vision Cloud
convolutional neural network(CNN)
modularization
visualization
kernel function