| 摘要: |
| 针对Gabbard等人发表在《Pyhsical Review Letters》上的文章“Matching Matched Filtering with Deep Networks for Gravitational-Wave Astronomy ”，提出了一种卷积神经网络优化模型。文章将卷积神经网络应用于引力波信号的识别，研究最大池化层参数对模型分类能力的影响，调整模型中超参数提升引力波信号分类的准确率；将优化后的网络结构与Gabbard 的卷积神经网络用于相同的模拟数据集，并在测试集上绘制了接受者操作特性曲线（Receiver Operating Characteristic curve，简称ROC 曲线）,计算了ROC 曲线下的面积；结果证明：相比于未优化的网络，此处的模型在ROC下的面积在数值上提高了0.025 4~0.032 6；同时，改变噪音的振幅，将两种方法应用于新的数据集上，结果同样证明，优化后网络效果更好，鲁棒性强。 |
| 关键词: 引力波天文学 深度学习方法 卷积神经网络 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Application of Optimization Model Based on Maximum Pooling Layer Parameters to Gravitational Wave Astronomy |
|
LUO Hua-mei
|
| Abstract: |
| n this paper, the convolutional neural network is applied to the identification of gravitational wave signals, the influence of the maximum pooling layer parameters on the classification ability of the model is studied, and the accuracy of the superparameters in the model to improve the classification of gravitational wave signals is adjusted. The optimized network structure and Gabbard′s convolutional neural network were used for the same simulation data set, and the Receiver Operating Characteristic curve (ROC curve) was plotted on the test set and the ROC curve area was calculated.The results show that our model has increased the area under ROC by 0.0254 to 0.0326 compared with the unoptimized network. At the same time, we also changed the amplitude of the noise and applied the two methods to the new data set. The results also prove that the network is better and more robust after optimization. |
| Key words: gravitational wave astronomy deep learning method convolutional neural network |
