近年来,卷积神经网络模型在文本分类中显示出了良好的应用前景,但该模型优势在于可以应用更深更广的卷积层来提取更丰富的语义特征,带来了昂贵的计算成本,并且在量级差异较大的数据集中不具备普适性。为了解决这一问题,提出了一种新型...近年来,卷积神经网络模型在文本分类中显示出了良好的应用前景,但该模型优势在于可以应用更深更广的卷积层来提取更丰富的语义特征,带来了昂贵的计算成本,并且在量级差异较大的数据集中不具备普适性。为了解决这一问题,提出了一种新型卷积网络结构,即用可分解两层卷积网络代替传统的文本卷积网络。一层词嵌入卷积层用来提取单词的词嵌入特征,另一层区域卷积层用来提取单词的上下文特征。模型在CPU上对多个数据集进行了测试,结果表明,该模型不仅降低了训练复杂度,在MR数据集上实现4 min 40 s的最短训练时间,而且在大数据集AG上准确率达到92.6,小数据集MR上达到83.0,这证明了模型在精度和鲁棒性方面都有良好的效果。此外,还进一步讨论了卷积核尺寸和数量对模型性能的影响,并且对以往的诸如SVM,CNN等经典文本分类模型做出了回顾,比较和总结,显现出了可分离卷积模型在算法复杂度和准确率方面的优越性。展开更多
The FOURIER transform is one of the most frequently used tools in signal analysis. A generalization of the Fourier transform-the fractional Fourier transform-has become a powerful tool for time-varying signal analys...The FOURIER transform is one of the most frequently used tools in signal analysis. A generalization of the Fourier transform-the fractional Fourier transform-has become a powerful tool for time-varying signal analysis. The mean square error(MSE) is used as design criteria to estimate signal. Wiener filter, which can be implement in O(NlogN) time, is suited for time-invariant degradation models. For time-variant and non-stationary processes, however, the linear estimate requires O(N 2 ). Filtering in fractional Fourier domains permits reduction of the error compared with ordinary Fourier domain filtering while requiring O(NlogN) implementation time. The blurred images that have several degradation models with different SNR are restored in the experiments. The results show that the method presented in this paper is valid and that the effect of restoration is improved as SNR is increased.展开更多
文摘近年来,卷积神经网络模型在文本分类中显示出了良好的应用前景,但该模型优势在于可以应用更深更广的卷积层来提取更丰富的语义特征,带来了昂贵的计算成本,并且在量级差异较大的数据集中不具备普适性。为了解决这一问题,提出了一种新型卷积网络结构,即用可分解两层卷积网络代替传统的文本卷积网络。一层词嵌入卷积层用来提取单词的词嵌入特征,另一层区域卷积层用来提取单词的上下文特征。模型在CPU上对多个数据集进行了测试,结果表明,该模型不仅降低了训练复杂度,在MR数据集上实现4 min 40 s的最短训练时间,而且在大数据集AG上准确率达到92.6,小数据集MR上达到83.0,这证明了模型在精度和鲁棒性方面都有良好的效果。此外,还进一步讨论了卷积核尺寸和数量对模型性能的影响,并且对以往的诸如SVM,CNN等经典文本分类模型做出了回顾,比较和总结,显现出了可分离卷积模型在算法复杂度和准确率方面的优越性。
文摘The FOURIER transform is one of the most frequently used tools in signal analysis. A generalization of the Fourier transform-the fractional Fourier transform-has become a powerful tool for time-varying signal analysis. The mean square error(MSE) is used as design criteria to estimate signal. Wiener filter, which can be implement in O(NlogN) time, is suited for time-invariant degradation models. For time-variant and non-stationary processes, however, the linear estimate requires O(N 2 ). Filtering in fractional Fourier domains permits reduction of the error compared with ordinary Fourier domain filtering while requiring O(NlogN) implementation time. The blurred images that have several degradation models with different SNR are restored in the experiments. The results show that the method presented in this paper is valid and that the effect of restoration is improved as SNR is increased.