检索结果-南通市图书馆

Computer Modeling in Engineering & Sciences 2023年第6期135卷 2155-2172页

作者： Xiaojun Zhu heming huang School of Computer Science Qinghai Normal UniversityXining810008China The State Key Laboratory of Tibetan Intelligent Information Processing and Application Xining810008China School of Electronic and Information Engineering Lanzhou City UniversityLanzhou730000China

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.

关键词： Speech enhancement generative adversarial networks hybrid penalty gated linear units multi-scale convolution

How could imperfect device properties influence the performances of spiking neural networks?

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2023年第8期66卷 233-242页

作者： Jingyang CHEN Zhihao WANG Tong WANG heming huang Zheyuan SHAO Zhe WANG Xin GUO State Key Laboratory of Material Processing and Die & Mould Technology School of Materials Science and EngineeringHuazhong University of Science and Technology

Spiking neural networks(SNNs) provide an efficient way to apply artificial intelligence systems on edge devices. Memristive devices with tunable conductance states can be used to emulate the functions of biological neurons and synapses and to build SNNs. In this work, fully-connected SNNs based on various memristive devices are constructed, and a hardware-compatible spike-timing-dependent plasticity(STDP)learning rule is applied to train the SNNs. Strategies are designed to suppress the overfitting problem and improve the performance of the SNNs in the case of a small training set. However, the properties of memristive devices are never perfect. The effects of imperfect device properties, e.g., asymmetric weight update, insufficient number of conductance states, and low on/off ratio, on the performance of the SNNs are elaborated.

关键词： oxides memristive devices device properties spiking neural networks neuromorphic computing

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

面向不平衡数据集的语音情感识别研究

计算机工程与应用 2024年第4期60卷 122-132页

作者：张会云黄鹤鸣青海师范大学计算机学院西宁810008 藏语智能语音信息处理及应用国家重点实验室西宁810008

样本平衡对机器学习至关重要,在不平衡数据集中,虽然某些类别的样本数量可能很少,但其重要性可能更高。研究了基于不平衡数据集的语音情感识别技术。在不同信噪比下采用不同噪声对不平衡基线数据集EMODB和IEMOCAP进行扩充,构建含噪数据... 详细信息

样本平衡对机器学习至关重要,在不平衡数据集中,虽然某些类别的样本数量可能很少,但其重要性可能更高。研究了基于不平衡数据集的语音情感识别技术。在不同信噪比下采用不同噪声对不平衡基线数据集EMODB和IEMOCAP进行扩充,构建含噪数据集EMODBM和IEMOCAPM;采用SMOTE、RandomOverSampler、SMOTEENN、ADASYN、TomekLinks以及SMOTETomek等6种技术对基线数据集和含噪数据集进行重采样,实现类别样本平衡;在基线数据集和扩充数据集上分别提取21维的低级描述符特征;采用新提出的模型MA-CapsNet验证重采样技术的有效性。实验表明,重采样后各类情感样本基本平衡,使模型的学习更公平、更客观,并且模型在重采样数据集上的鲁棒性更好。

关键词：语音情感识别重采样胶囊网络数据扩充

Uncovering recent progress in nanostructured light-emitters for information and communication technologies

在线全文

学校读者我要写书评

暂无评论

Light(Science & Applications) 2021年第9期10卷 1571-1587页

作者： Frédéric Grillot Jianan Duan Bozhang Dong heming huang LTCI Institut Polytechnique de ParisTélécom Paris19 place Marguerite Perey91120PalaiseauFrance Center for High Technology Materials University of New-Mexico1313 Goddard St SEAlbuquerqueNM87106USA State Key Laboratory on Tunable Laser Technology School of Electronic and Information EngineeringHarbin Institute of TechnologyShenzhen518055China

Semiconductor nanostructures with low dimensionality like quantum dots and quantum dashes are one of the best attractive and heuristic solutions for achieving high performance photonic devices.When one or more spatial dimensions of the nanocrystal approach the de Broglie wavelength,nanoscale size effects create a spatial quantization of carriers leading to a complete discretization of energy levels along with additional quantum phenomena like entangled-photon generation or squeezed states of light among others.This article reviews our recent findings and prospects on nanostructure based light emitters where active region is made with quantum-dot and quantum-dash nanostructures.Many applications ranging from silicon-based integrated technologies to quantum information systems rely on the utilization of such laser sources.Here,we link the material and fundamental properties with the device physics.For this purpose,spectral linewidth,polarization anisotropy,optical nonlinearities as well as microwave,dynamic and nonlinear properties are closely examined.The paper focuses on photonic devices grown on native substrates(InP and GaAs)as well as those heterogeneously and epitaxially grown on silicon substrate.This research pipelines the most exciting recent innovation developed around light emitters using nanostructures as gain media and highlights the importance of nanotechnologies on industry and society especially for shaping the future information and communication society.

关键词： emitter communication polarization

Inhibiting p38 mitogen-activated protein kinase attenuates cerebral ischemic injury in Swedish mutant amyloid precursor protein transgenic mice

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Neural Regeneration Research 2012年第14期7卷 1088-1094页

作者： Liangyu Zou Haiyan Qin Yitao He heming huang Yi Lu Xiaofan Chu Department of Neurology Shenzhen People's HospitalSecond Clinical CollegeJinan UniversityShenzhen 518020Guangdong ProvinceChina

Cerebral ischemia was induced using photothrombosis 1 hour after intraperitoneal injection of the p38 mitogen-activated protein kinase （MAPK） inhibitor $B239063 into Swedish mutant amyloid precursor protein （APP/SWE） transgenic and non-transgenic mice. The number of surviving neurons in the penumbra was quantified using Nissl staining, and the activity of p38 MAPKs was measured by western blotting. The number of surviving neurons in the penumbra was significantly reduced in APP/SWE transgenic mice compared with non-transgenic controls 7 days after cerebral ischemia, but the activity of p38 MAPKs was significantly elevated compared with the non-ischemic hemisphere in the APP/SWE transgenic mice. SB239063 prevented these changes. The APP/SWE mutation exacerbated ischemic brain injury, and this could be alleviated by inhibiting p38 MAPK activity.

关键词： cerebral ischemia amyloid precursor protein transgenic Alzheimer＇s disease p38mitogen-activated protein kinase SB239063 neural regeneration

在线全文

学校读者我要写书评

暂无评论

融合多尺度特征与上下文信息的语音增强方法

计算机工程 2024年第6期50卷 138-147页

作者：更藏措毛黄鹤鸣杨毅杰青海师范大学计算机学院青海西宁810008 藏语智能信息处理及应用国家重点实验室青海西宁810000

在语音增强中,常用自编码器结构自动提取特征,但这样得到的特征单一或者冗余且不能较好地捕获语音信号的上下文依赖关系。因此,提出一种融合多尺度特征和上下文信息的语音增强方法MSF-CI。首先,利用多尺度卷积块提取语音信号的多尺度特... 详细信息

在语音增强中,常用自编码器结构自动提取特征,但这样得到的特征单一或者冗余且不能较好地捕获语音信号的上下文依赖关系。因此,提出一种融合多尺度特征和上下文信息的语音增强方法MSF-CI。首先,利用多尺度卷积块提取语音信号的多尺度特征,解决特征单一问题;其次,利用注意力机制关注所提取特征的空间与通道关键信息,解决特征冗余问题;最后,使用门控卷积循环神经网络学习语音信号中跨度较长的上下文依赖关系,并通过门控线性单元提高该网络的非线性学习能力,从而提高模型的泛化性。实验结果表明,MSF-CI在低信噪比和不同噪声环境下增强语音信号的语音感知质量、短时客观可懂度等多个指标上均优于GRN、DPT-FSNet、U-Net等同类的单通道语音增强模型。在信噪比为0 dB时,该方法的平均语音感知质量和平均语音客观可懂度达到1.49和0.761。在构建的安多藏语语料库上验证模型的泛化性,平均语音感知质量和平均语音客观可懂度相对于噪声提高了20.7%和11.3%,MSF-CI模型不仅可以提升语音的质量与可理解度,而且具有较优的泛化性。

关键词：语音增强多尺度特征注意力机制门控卷积循环神经网络对数能量谱

Dynamic and nonlinear properties of epitaxial quantum dot lasers on silicon for isolator-free integration

在线全文

学校读者我要写书评

暂无评论

Photonics Research 2019年第11期7卷 1222-1228页

作者： Jianan Duan heming huang Bozhang Dong Justin C.Norman Zeyu Zhang John E.Bowers Frederic Grillot LTCI Telecom ParisInstitut Polytechnique de Paris46 rue Barrault75013 ParisFrance Materials Department University of CaliforniaSanta BarbaraCalifornia 93106USA Department of Electrical and Computer Engineering University of CaliforniaSanta BarbaraCalifornia 93106USA Institute for Energy Efficiency University of CaliforniaSanta BarbaraCalifornia 93106USA Center for High Technology Materials University of New MexicoAlbuquerqueNew Mexico 87106USA

This work investigates the dynamic and nonlinear properties of quantum dot(QD) lasers directly grown on silicon with a view to isolator-free applications. Among them, the chirp parameter, also named the αHfactor,is featured through a thermally insensitive method analyzing the residual side-mode dynamics under optical injection locking. The αHat threshold is found as low as 0.32. Then, the nonlinear gain is investigated from the gain compression factor viewpoint. The latter is found higher for epitaxial QD lasers on silicon than that in heterogeneously integrated quantum well(QW) devices on silicon. Despite that, the power dependence of the αHdoes not lead to a large increase of the chirp coefficient above the laser’s threshold at higher bias. This effect is confirmed from an analytical model and attributed to the strong lasing emission of the ground-state transition, which transforms into a critical feedback level as high as-6.5 d B, which is ~19 d B higher than a comparable QW laser.Finally, the intensity noise analysis confirms that QD lasers are overdamped oscillators with damping frequencies as large as 33 GHz. Altogether, these features contribute to fundamentally enhancing the reflection insensitivity of the epitaxial QD lasers. This last feature is unveiled by the 10 Gbit/s error-free high-speed transmission experiments. Overall, we believe that this work is of paramount importance for future isolator-free photonics technologies and cost-efficient high-speed transmission systems.

关键词： quantum nonlinear epitaxial

Reflection sensitivity of dual-state quantum dot lasers

在线全文

学校读者我要写书评

暂无评论

Photonics Research 2023年第10期11卷 1713-1722页

作者： ZHIYONG JIN heming huang YUEGUANG ZHOU SHIYUAN ZHAO SHIHAO DING CHENG WANG YONG YAO XIAOCHUAN XU FRÉDÉRIC GRILLOT JIANAN DUAN State Key Laboratory on Tunable Laser Technology School of Electronic and Information EngineeringHarbin Institute of TechnologyShenzhen 518055China LTCI Telecom ParisInstitut Polytechnique de Paris91120 PalaiseauFrance DTU Electro Department of Electrical and Photonics EngineeringTechnical University of Denmark2800 LyngbyDenmark School of Information Science and Technology ShanghaiTech UniversityShanghai 201210China Center for High Technology Materials The University of New-MexicoAlbuquerqueNew Mexico 87106USA

This work experimentally and theoretically demonstrates the effect of excited state lasing on the reflection sensitivity of dual-state quantum dot lasers,showing that the laser exhibits higher sensitivity to external optical feedback when reaching the excited state lasing threshold.This sensitivity can be degraded by increasing the excited-to-ground-state energy separation,which results in a high excited-to-ground-state threshold ratio.In addition,the occurrence of excited state lasing decreases the damping factor and increases the linewidth enhancement factor,which leads to a low critical feedback level.These findings illuminate a path to fabricate reflectioninsensitive quantum dot lasers for isolator-free photonic integrated circuits.

关键词： state excited quantum

Four-wave mixing in 1.3 μm epitaxial quantum dot lasers directly grown on silicon

在线全文

学校读者我要写书评

暂无评论

Photonics Research 2022年第5期10卷 1264-1270页

作者： JIANAN DUAN BOZHANG DONG WENG W.CHOW heming huang SHIHAO DING SONGTAO LIU JUSTIN C.NORMAN JOHN E.BOWERS FREDEIC GRILLOT LTCI Telecom ParisInstitut Polytechnique de Pahs91120 PalaiseauFrance Sandia National Laboratories AlbuquerqueNew Mexico 87185USA Institute for Energy Efficiency University of California Santa BarbaraSanta BarbaraCalifornia 93106USA Center for High Technology Materials University of New-MexicoAlbuquerqueNew Mexico 87106USA State Key Laboratory on Tunable Laser Technology School of Electronic and Information EngineeringHarbin Institute of TechnologyShenzhen 518055China Ayar Labs Santa ClaraCalifornia 95054USA Quintessent Inc.GoletaCalifornia 93117USA

This work compares the four-wave mixing(FWM)effect in epitaxial quantum dot(QD)lasers grown on silicon with quantum well(QW)lasers.A comparison of theory and experiment results shows that the measured FWM coefficient is in good agreement with theoretical predictions.The gain in signal power is higher for p-doped QD lasers than for undoped lasers,despite the same FWM coefficient.Owing to the near-zero linewidth enhancement factor,QD lasers exhibit FWM coefficients and conversion efficiency that are more than one order of magnitude higher than those of QW lasers.Thus,this leads to self-mode locking in QD lasers.These findings are useful for developing on-chip sources for photonic integrated circuits on silicon.

关键词： lasers quantum mixing

基于无监督学习和监督学习的抽取式文本摘要综述

在线全文

学校读者我要写书评

暂无评论

计算机应用 2024年第4期44卷 1035-1048页

作者：夏吾吉黄鹤鸣更藏措毛范玉涛青海师范大学计算机学院西宁810008 藏语智能信息处理及应用国家重点实验室(青海师范大学) 西宁810008

相较于生成式摘要方法,抽取式摘要方法简单易行、可读性强,使用范围广。目前,抽取式摘要方法综述文献仅对特定的某个方法或领域进行分析综述,缺乏多方面、多语种的系统性综述,因此探讨文本摘要生成任务的内涵,通过系统梳理和提炼现有的... 详细信息

相较于生成式摘要方法,抽取式摘要方法简单易行、可读性强,使用范围广。目前,抽取式摘要方法综述文献仅对特定的某个方法或领域进行分析综述,缺乏多方面、多语种的系统性综述,因此探讨文本摘要生成任务的内涵,通过系统梳理和提炼现有的相关文献,对无监督学习和监督学习的抽取式文本摘要技术进行多维度、全方位的分析。首先,回顾文本摘要技术的发展,分析不同的抽取式文本摘要方法,主要包括基于规则、词频-逆文件概率(TFIDF)、中心性方法、潜在语义、深度学习、图排序、特征工程和预训练学习等,并对比不同方法的差异;其次,详细介绍不同语种文本摘要生成的常用数据集和主流的评价指标,通过不同的实验指标对相同数据集上的方法进行比较;最后,指出当前抽取式文本摘要研究中存在的主要问题和挑战,并提出具体的解决思路和未来发展趋势。

关键词：抽取式摘要无监督学习监督学习数据集评价指标