Recently,speech enhancement methods based on Generative Adversarial Networks have achieved good performance in time-domain noisy signals.However,the training of Generative Adversarial Networks has such problems as con...Recently,speech enhancement methods based on Generative Adversarial Networks have achieved good performance in time-domain noisy signals.However,the training of Generative Adversarial Networks has such problems as convergence difficulty,model collapse,etc.In this work,an end-to-end speech enhancement model based on Wasserstein Generative Adversarial Networks is proposed,and some improvements have been made in order to get faster convergence speed and better generated speech quality.Specifically,in the generator coding part,each convolution layer adopts different convolution kernel sizes to conduct convolution operations for obtaining speech coding information from multiple scales;a gated linear unit is introduced to alleviate the vanishing gradient problem with the increase of network depth;the gradient penalty of the discriminator is replaced with spectral normalization to accelerate the convergence rate of themodel;a hybrid penalty termcomposed of L1 regularization and a scale-invariant signal-to-distortion ratio is introduced into the loss function of the generator to improve the quality of generated speech.The experimental results on both TIMIT corpus and Tibetan corpus show that the proposed model improves the speech quality significantly and accelerates the convergence speed of the model.展开更多
Linear optical quantum Fredkin gate can be applied to quantum computing and quantum multi-user communication networks. In the existing linear optical scheme, two single photon detectors （SPDs） are used to herald the...Linear optical quantum Fredkin gate can be applied to quantum computing and quantum multi-user communication networks. In the existing linear optical scheme, two single photon detectors （SPDs） are used to herald the success of the quantum Fredkin gate while they have no photon count. But analysis results show that for non-perfect SPD, the lower the detector efficiency, the higher the heralded success rate by this scheme is. We propose an improved linear optical quantum Fredkin gate by designing a new heralding scheme with an auxiliary qubit and only one SPD, in which the higher the detection efficiency of the heralding detector, the higher the success rate of the gate is. The new heralding scheme can also work efficiently under a non-ideal single photon source. Based on this quantum Fredkin gate, large-scale quantum switching networks can be built. As an example, a quantum Bene~ network is shown in which only one SPD is used.展开更多
According to the study of super-resolution range-gated system, we proposed an improved system with linear plus detects. And a range function is derived by considering the shot effect noise and dark current noise. The ...According to the study of super-resolution range-gated system, we proposed an improved system with linear plus detects. And a range function is derived by considering the shot effect noise and dark current noise. The simulation shows that the improved system has a good range accuracy capability.展开更多
The paper reveals the relation between the linear AND-OR gate and the emitter function logic. With theoretic calculation and PSPICE simulation, the paper proves that the linear AND-OR gates can work at super-high-spee...The paper reveals the relation between the linear AND-OR gate and the emitter function logic. With theoretic calculation and PSPICE simulation, the paper proves that the linear AND-OR gates can work at super-high-speed and can be multi-cascaded. On the basis of analyzing the high-speed switch units which coordinate with linear AND-OR gates, two kinds of emitter coupled logic circuits are designed. The paper also discusses the design principles of super-high-speed digital circuits, and some examples of combinational and sequential circuits using linear AND-OR gate are given.展开更多
水质指标具有多元相关性、时序性和非线性的特点,为有效预测河流水质变化,针对水质数据存在缺失和异常的问题,提出基于灰色关联分析-门控循环单元(Grey Relational Analysis-Gated Recurrent Unit, GRA-GRU)的水质预测模型。以淮河流域...水质指标具有多元相关性、时序性和非线性的特点,为有效预测河流水质变化,针对水质数据存在缺失和异常的问题,提出基于灰色关联分析-门控循环单元(Grey Relational Analysis-Gated Recurrent Unit, GRA-GRU)的水质预测模型。以淮河流域水质数据为样本,使用线性插值修补缺失数据和剔除的异常数据。使用灰色关联分析计算不同水质指标间的相关性,选择高相关性的水质指标以确定输入变量,并使用门控循环单元(Gated Recurrent Unit, GRU)预测不同的水质指标。将GRA-GRU的预测结果与反向传播神经网络(Back Propagation Neural Network, BPNN)、循环神经网络(Recurrent Neural Network, RNN)、长短期记忆神经网络(Long Short Term Memory, LSTM)、GRU及灰色关联分析-长短期记忆神经网络(Grey Relational Analysis-Long Short Term Memory, GRA-LSTM)进行对比分析,结果显示GRA-GRU在不同水质指标预测上具有较好的适应性,可以有效降低预测误差。其中,与其他模型相比,GRA-GRU预测的化学需氧量在均方根误差上分别降低了3.617%、0.681%、0.478%、1.505%和0.471%。展开更多
现有分阶段解码的实体关系抽取模型仍存在着阶段间特征融合不充分的问题,会增大曝光偏差对抽取性能的影响。为此,提出一种双关系预测和特征融合的实体关系抽取模型(entity relation extraction model with dual relation prediction and...现有分阶段解码的实体关系抽取模型仍存在着阶段间特征融合不充分的问题,会增大曝光偏差对抽取性能的影响。为此,提出一种双关系预测和特征融合的实体关系抽取模型(entity relation extraction model with dual relation prediction and feature fusion,DRPFF),该模型使用预训练的基于Transformer的双向编码表示模型(bidirectional encoder representation from transformers,BERT)对文本进行编码,并设计两阶段的双关系预测结构来减少抽取过程中错误三元组的生成。在阶段间通过门控线性单元(gated linear unit,GLU)和条件层规范化(conditional layer normalization,CLN)组合的结构来更好地融合实体之间的特征。在NYT和WebNLG这2个公开数据集上的试验结果表明,该模型相较于基线方法取得了更好的效果。展开更多
基金supported by the National Science Foundation under Grant No.62066039.
文摘Recently,speech enhancement methods based on Generative Adversarial Networks have achieved good performance in time-domain noisy signals.However,the training of Generative Adversarial Networks has such problems as convergence difficulty,model collapse,etc.In this work,an end-to-end speech enhancement model based on Wasserstein Generative Adversarial Networks is proposed,and some improvements have been made in order to get faster convergence speed and better generated speech quality.Specifically,in the generator coding part,each convolution layer adopts different convolution kernel sizes to conduct convolution operations for obtaining speech coding information from multiple scales;a gated linear unit is introduced to alleviate the vanishing gradient problem with the increase of network depth;the gradient penalty of the discriminator is replaced with spectral normalization to accelerate the convergence rate of themodel;a hybrid penalty termcomposed of L1 regularization and a scale-invariant signal-to-distortion ratio is introduced into the loss function of the generator to improve the quality of generated speech.The experimental results on both TIMIT corpus and Tibetan corpus show that the proposed model improves the speech quality significantly and accelerates the convergence speed of the model.
基金Project supported by the National Natural Science Foundation of China(Grant Nos.61372076 and 61301171)the 111 Project(Grant No.B08038)the Fundamental Research Funds for the Central Universities,China(Grant Nos.K5051301059 and K5051201021)
文摘Linear optical quantum Fredkin gate can be applied to quantum computing and quantum multi-user communication networks. In the existing linear optical scheme, two single photon detectors （SPDs） are used to herald the success of the quantum Fredkin gate while they have no photon count. But analysis results show that for non-perfect SPD, the lower the detector efficiency, the higher the heralded success rate by this scheme is. We propose an improved linear optical quantum Fredkin gate by designing a new heralding scheme with an auxiliary qubit and only one SPD, in which the higher the detection efficiency of the heralding detector, the higher the success rate of the gate is. The new heralding scheme can also work efficiently under a non-ideal single photon source. Based on this quantum Fredkin gate, large-scale quantum switching networks can be built. As an example, a quantum Bene~ network is shown in which only one SPD is used.
文摘According to the study of super-resolution range-gated system, we proposed an improved system with linear plus detects. And a range function is derived by considering the shot effect noise and dark current noise. The simulation shows that the improved system has a good range accuracy capability.
基金Supported by the National Natural Science Foundation of China
文摘The paper reveals the relation between the linear AND-OR gate and the emitter function logic. With theoretic calculation and PSPICE simulation, the paper proves that the linear AND-OR gates can work at super-high-speed and can be multi-cascaded. On the basis of analyzing the high-speed switch units which coordinate with linear AND-OR gates, two kinds of emitter coupled logic circuits are designed. The paper also discusses the design principles of super-high-speed digital circuits, and some examples of combinational and sequential circuits using linear AND-OR gate are given.
文摘水质指标具有多元相关性、时序性和非线性的特点,为有效预测河流水质变化,针对水质数据存在缺失和异常的问题,提出基于灰色关联分析-门控循环单元(Grey Relational Analysis-Gated Recurrent Unit, GRA-GRU)的水质预测模型。以淮河流域水质数据为样本,使用线性插值修补缺失数据和剔除的异常数据。使用灰色关联分析计算不同水质指标间的相关性,选择高相关性的水质指标以确定输入变量,并使用门控循环单元(Gated Recurrent Unit, GRU)预测不同的水质指标。将GRA-GRU的预测结果与反向传播神经网络(Back Propagation Neural Network, BPNN)、循环神经网络(Recurrent Neural Network, RNN)、长短期记忆神经网络(Long Short Term Memory, LSTM)、GRU及灰色关联分析-长短期记忆神经网络(Grey Relational Analysis-Long Short Term Memory, GRA-LSTM)进行对比分析,结果显示GRA-GRU在不同水质指标预测上具有较好的适应性,可以有效降低预测误差。其中,与其他模型相比,GRA-GRU预测的化学需氧量在均方根误差上分别降低了3.617%、0.681%、0.478%、1.505%和0.471%。
文摘现有分阶段解码的实体关系抽取模型仍存在着阶段间特征融合不充分的问题,会增大曝光偏差对抽取性能的影响。为此,提出一种双关系预测和特征融合的实体关系抽取模型(entity relation extraction model with dual relation prediction and feature fusion,DRPFF),该模型使用预训练的基于Transformer的双向编码表示模型(bidirectional encoder representation from transformers,BERT)对文本进行编码,并设计两阶段的双关系预测结构来减少抽取过程中错误三元组的生成。在阶段间通过门控线性单元(gated linear unit,GLU)和条件层规范化(conditional layer normalization,CLN)组合的结构来更好地融合实体之间的特征。在NYT和WebNLG这2个公开数据集上的试验结果表明,该模型相较于基线方法取得了更好的效果。
基金the National Natural Science Foundation of China(62304252)the Youth Innovation Promotion Association of Chinese Academy Sciences(CAS)and IMECAS-HKUST-Joint Laboratory of Microelectronics。
