Randomized weights neural networks have fast learning speed and good generalization performance with one single hidden layer structure. Input weighs of the hidden layer are produced randomly. By employing certain acti...Randomized weights neural networks have fast learning speed and good generalization performance with one single hidden layer structure. Input weighs of the hidden layer are produced randomly. By employing certain activation function, outputs of the hidden layer are calculated with some randomization. Output weights are computed using pseudo inverse. Mutual information can be used to measure mutual dependence of two variables quantitatively based on the probability theory. In this paper, these hidden layer’s outputs that relate to prediction variable closely are selected with the simple mutual information based feature selection method. These hidden nodes with high mutual information values are maintained as a new hidden layer. Thus, the size of the hidden layer is reduced. The new hidden layer’s output weights are learned with the pseudo inverse method. The proposed method is compared with the original randomized algorithms using concrete compressive strength benchmark dataset.展开更多
命名实体识别是知识抽取中的重要任务之一,为了更有效地利用词典匹配信息,提出了基于匹配词权重优化的中文命名实体识别模型。首先利用与训练模型和分词工具获得每个字符的向量表示和词性标注;然后在词典中匹配潜在词组,跟据匹配词词频...命名实体识别是知识抽取中的重要任务之一,为了更有效地利用词典匹配信息,提出了基于匹配词权重优化的中文命名实体识别模型。首先利用与训练模型和分词工具获得每个字符的向量表示和词性标注;然后在词典中匹配潜在词组,跟据匹配词词频和文档计数的优化权重对词组加权,结合字符向量得到字符的多特征融合表示;最后使用双向长短期记忆网络(Bi-directional Long-Short Term Memory,Bi-LSTM)网络进行训练,使用条件随机场(Conditional Random Field,CRF)完成标签推理得到识别实体。试验结果表明,该模型在Resume和影视-音乐-书籍数据集上的F1值分别达到了95.55%和85.39%,有效地提高了中文命名实体识别效果。展开更多
为有效提高地震数据信噪比,通过卷积神经网络(convolutional neural network,CNN)的方法研究了地震勘探数据去除随机噪声问题。该方法包含17个卷积层,使用线性整流(rectified linear unit,ReLU)激活函数避免梯度消失,使用批量标准化(bat...为有效提高地震数据信噪比,通过卷积神经网络(convolutional neural network,CNN)的方法研究了地震勘探数据去除随机噪声问题。该方法包含17个卷积层,使用线性整流(rectified linear unit,ReLU)激活函数避免梯度消失,使用批量标准化(batch normalization,BN)提高网络的泛化能力。所构建的网络应用残差学习策略,即输入为含噪地震正演叠前数据,输出为CNN网络学习获得的随机噪声。然后从地震记录中减去网络预测的噪声数据,从而达到去除随机噪声的目的。同时,根据地震勘探数据振幅随探测时间衰减的规律,在网络训练过程中进行深度加权,使得CNN对于深部噪声的学习效果更好。网络在PyTorch框架下训练,应用图形处理器并行计算可以有效提高网络训练速度。利用训练好的网络进行去噪实验,结果表明与传统的时空域预测滤波法相比,该网络能更好地压制随机噪声。可见针对地震勘探数据,CNN能够有效提取含噪数据中的噪声信息,证明了该方法在去除随机噪声方面的合理性与有效性。展开更多
文摘Randomized weights neural networks have fast learning speed and good generalization performance with one single hidden layer structure. Input weighs of the hidden layer are produced randomly. By employing certain activation function, outputs of the hidden layer are calculated with some randomization. Output weights are computed using pseudo inverse. Mutual information can be used to measure mutual dependence of two variables quantitatively based on the probability theory. In this paper, these hidden layer’s outputs that relate to prediction variable closely are selected with the simple mutual information based feature selection method. These hidden nodes with high mutual information values are maintained as a new hidden layer. Thus, the size of the hidden layer is reduced. The new hidden layer’s output weights are learned with the pseudo inverse method. The proposed method is compared with the original randomized algorithms using concrete compressive strength benchmark dataset.
文摘命名实体识别是知识抽取中的重要任务之一,为了更有效地利用词典匹配信息,提出了基于匹配词权重优化的中文命名实体识别模型。首先利用与训练模型和分词工具获得每个字符的向量表示和词性标注;然后在词典中匹配潜在词组,跟据匹配词词频和文档计数的优化权重对词组加权,结合字符向量得到字符的多特征融合表示;最后使用双向长短期记忆网络(Bi-directional Long-Short Term Memory,Bi-LSTM)网络进行训练,使用条件随机场(Conditional Random Field,CRF)完成标签推理得到识别实体。试验结果表明,该模型在Resume和影视-音乐-书籍数据集上的F1值分别达到了95.55%和85.39%,有效地提高了中文命名实体识别效果。
文摘针对污水处理复杂系统中关键水质参数生化需氧量(biochemical oxygen demand,BOD)难以准确实时预测的问题,在分析污水处理过程相关影响因素的基础上,提出一种基于敏感度分析法的自组织随机权神经网络(selforganizing neural network with random weights,SONNRW)软测量方法.该方法首先通过机理分析选取原始辅助变量,经过数据预处理,之后采用主元分析法对辅助变量进行精选,作为SONNRW的输入变量进行污水处理关键水质参数BOD的预测.SONNRW算法利用隐含层节点输出及其权值向量计算该隐含层节点对于残差的敏感度,根据敏感度大小对网络隐含层节点进行排序,删除敏感度较低的隐含层节点即冗余点.仿真结果表明:该软测量方法对水质参数BOD的预测精度高、实时性好、模型结构稳定,能够用于污水水质的在线预测.
