检索结果-南通市图书馆

Binary neural networks for speech recognition

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Frontiers of Information Technology & Electronic engineering 2019年第5期20卷 701-715页

作者： Yan-min QIAN Xu XIANG key laboratory of shanghai education commission for intelligent interaction and cognitive engineering Shanghai Jiao Tong University Shanghai 200240 China SpeechLab Department of Computer Science and Engineering Shanghai Jiao Tong University Shanghai 200240 China

Recently, deep neural networks(DNNs) significantly outperform Gaussian mixture models in acoustic modeling for speech recognition. However, the substantial increase in computational load during the inference stage makes deep models difficult to directly deploy on low-power embedded devices. To alleviate this issue,structure sparseness and low precision fixed-point quantization have been applied widely. In this work, binary neural networks for speech recognition are developed to reduce the computational cost during the inference stage.A fast implementation of binary matrix multiplication is introduced. On modern central processing unit(CPU)and graphics processing unit(GPU) architectures, a 5–7 times speedup compared with full precision floatingpoint matrix multiplication can be achieved in real applications. Several kinds of binary neural networks and related model optimization algorithms are developed for large vocabulary continuous speech recognition acoustic modeling. In addition, to improve the accuracy of binary models, knowledge distillation from the normal full precision floating-point model to the compressed binary model is explored. Experiments on the standard Switchboard speech recognition task show that the proposed binary neural networks can deliver 3–4 times speedup over the normal full precision deep models. With the knowledge distillation from the normal floating-point models, the binary DNNs or binary convolutional neural networks(CNNs) can restrict the word error rate(WER) degradation to within 15.0%,compared to the normal full precision floating-point DNNs or CNNs, respectively. Particularly for the binary CNN with binarization only on the convolutional layers, the WER degradation is very small and is almost negligible with the proposed approach.

关键词： Speech recognition Binary neural net works Binary matrix multiplication Knowledge distillation Population count

WEDeepT3: predicting type Ⅲ secreted effectors based on word embedding and deep learning

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Quantitative Biology 2019年第4期7卷 293-301页

作者： Xiaofeng Fu Yang Yang Department of Computer Science and engineering Shanghai Jiao Tbng Universityand Key Laboratory of Shanghai Education Commission for Intelligent Interaction and Cognitive EngineeringShanghai 200240China

Background:The type Ⅲ secreted effectors(T3SEs)are one of the indispensable proteins in the growth and reproduction of Gram-negative bacteria.In particular,the pathogenesis of Gram-negative bacteria depends on the type Ⅲ secreted effectors,and by injecting T3SEs into a host cell,the host cell's immunity can be destroyed.The high diversity of T3SE sequences and the lack of defined secretion signals make it difficult to identify and predict.Moreover,the related study of the pathological system associated with T3SE remains a hot topic in bioinformatics.Some computational tools have been developed to meet the growing demand for the recognition of T3SEs and the studies of type Ⅲ secretion systems(T3SS).Although these tools can help biological experiments in certain procedures,there is still room for improvement,even for the current best model,as the existing methods adopt handdesigned feature and traditional machine learning methods.Methods:In this study,we propose a powerful predictor based on deep learning methods,called WEDeepT3.Our work consists mainly of three key steps.First,we train word embedding vectors for protein sequences in a large-scale amino acid sequence database.Second,we combine the word vectors with traditional features extracted from protein sequences,like PSSM,to construct a more comprehensive feature representation.Finally,we construct a deep neural network model in the prediction of type Ⅲ secreted effectors.Results:The feature representation of WEDeepT3 consists of both word embedding and position-specific features.Working together with convolutional neural networks,the new model achieves superior performance to the state-ofthe-art methods,demonstrating the effectiveness of the new feature representation and the powerful learning ability of deep models.Conclusion:WEDeepT3 exploits both semantic information of Ar-mer fragments and evolutional information of protein sequences to accurately difYerentiate between T3SEs and non-T3SEs.WEDeepT3 is available at bcmi.***.edu.cn/~yangyang/WEDeepT3.html.

关键词： typeⅢsecreted effectors word2vector PSSM feature representation

Evolvable dialogue state tracking for statistical dialogue management

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

Frontiers of Computer Science 2016年第2期10卷 201-215页

作者： Kai YU Lu CHEN Kai SUN Qizhe XIE Su ZHU key laboratory of shanghai education commission for intelligent interaction and cognitive engineeringSpeechLab Department of Computer Science and Engineering Shanghai Jiao Tong University Shanghai 200240 China Department of Computer Science Cornell University Ithaca NY 14853 USA

Statistical dialogue management is the core of cognitive spoken dialogue systems （SDS） and has attracted great research interest. In recent years, SDS with the ability of evolution is of particular interest and becomes the cuttingedge of SDS research. Dialogue state tracking （DST） is a process to estimate the distribution of the dialogue states at each dialogue turn, given the previous interaction history. It plays an important role in statistical dialogue management. To provide a common testbed for advancing the research of DST, international DST challenges （DSTC） have been organised and well-attended by major SDS groups in the world. This paper reviews recent progresses on rule-based and statistical approaches during the challenges. In particular, this paper is focused on evolvable DST approaches for dialogue domain extension. The two primary aspects for evolution, semantic parsing and tracker, are discussed. Semantic enhancement and a DST framework which bridges rule-based and statistical models are introduced in detail. By effectively incorporating prior knowledge of dialogue state transition and the ability of being data-driven, the new framework supports reliable domain extension with little data and can continuously improve with more data available. This makes it excellent candidate for DST evolution. Experiments show that the evolvable DST approaches can achieve the state-of-the-art performance and outperform all previously submitted trackers in the third DSTC.

关键词： dialogue management domain extension,evolvable dialogue state tracking parser tracker

Human iPS Cells Loaded with MnO2-Based Nanoprobes for Photodynamic and Simultaneous Enhanced Immunotherapy Against Cancer

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

Nano-Micro Letters 2020年第10期12卷 36-52页

作者： Yanlei Liu Jingxing Yang Bin Liu Wen Cao Jingpu Zhang Yuming Yang Lijun Ma Jesus Martinez de la Fuente Jie Song Jian Ni Chunfu Zhang Daxiang Cui Institute of Nano Biomedicine and engineering Key Laboratory for Thin Film and Microfabrication of Ministration of EducationShanghai Engineering Research Centre for Intelligent Diagnosis and Treatment InstrumentDepartment of Instrument Science and EngineeringSchool of Electronic Information and Electrical EngineeringShanghai Jiao Tong University800 Dongchuan RoadShanghai 200240People’s Republic of China National engineering Center for Nanotechnology Shanghai 200240People’s Republic of China Department of Nuclear Medicine Rui Jin HospitalSchool of Biomedical EngineeringShanghai Jiao Tong UniversityShanghai 200240People’s Republic of China Scientific Research Center Shanghai Public Health Clinical CenterFudan UniversityShanghai 201508People’s Republic of China Tongren Hospital Shanghai Jiao Tong University School of Medicine1111 XianXia RoadShanghai 200336People’s Republic of China

How to trigger strong anti-tumor immune responses has become a focus for tumor therapy.Here,we report the human-induced pluripotent stem cells(iPSs)to deliver MnO2@Ce6 nanoprobes into tumors for simultaneous photodynamic therapy(PDT)and enhanced immunotherapy.Ce6 photosensitizer was attached on manganese dioxide(MnO2)nanoparticles,and resultant MnO2@Ce6 nanoprobes were delivered into mitomycin-treated iPSs to form iPS-MnO2@Ce6 nanoprobes.The iPS-MnO2@Ce6 actively targeted in vivo tumors,the acidic microenvironment triggered interaction between MnO2 and H2O2,released large quantities of oxygen,alleviated hypoxia in tumor.Upon PDT,singlet oxygen formed,broken iPSs released tumor-shared antigens,which evoked an intensive innate and adaptive immune response against the tumor,improving dendritic cells matured,effector T cells,and natural killer cells were activated.Meanwhile,regulatory T cells were reduced,and then the immune response induced by iPS-MnO2@Ce6 was markedly stronger than the immune reaction induced by MnO2@Ce6(P<0.05).The iPS-MnO2@Ce6 markedly inhibited tumor growth and metastasis and reduced mortality in mice models with tumor.Human iPS s loaded with MnO2-based nanoprobes are a promising strategy for simultaneous PDT and enhanced immunotherapy against tumor and own clinical translational prospect.

关键词： Human iPS MnO2@Ce6 nanoprobes Photodynamic therapy Immunotherapy Cancer

Robust flocking for non-identical second-order nonlinear multi-agent systems

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Autonomous intelligent Systems 2021年第1期1卷 81-90页

作者： Xiuxian Li Housheng Su Li Li Department of Control Science and engineering College of Electronics and Information Engineeringand Shanghai Research Institute for Intelligent Autonomous SystemsTongji UniversityShanghai 200092China Institute for Advanced Study Tongji UniversityShanghai 200092China School of Automation Image Processing and Intelligent Control Key Laboratory of Education Ministry of ChinaHuazhong University of Science and TechnologyLuoyu Road 1037Wuhan 430074China

This paper investigates the robust flocking problem for second-order nonlinear systems with a leader and external disturbances.In contrast with most of second-order systems in the literature,the intrinsic dynamics here are nonlinear and non-identical that depend not only on the velocity but also on the position,which is more realistic.Moreover,the interaction topology is undirected and switching.Provided that the leader’s velocity may be constant or time-varying,two distributed flocking control laws have been proposed for two cases to make the differences of the velocities between all followers and the leader approach to zero asymptotically.The proposed distributed flocking control laws are both model-independent which results in the effectiveness of the controllers to cope with the different intrinsic dynamics of the followers and the leader under some assumptions on boundedness of several states.An example is given to illustrate the validity of the theoretical results.

关键词： Flocking Multi-agent systems Second-order nonlinear systems Non-identical dynamics

Fusion Model for Speech Emotion Recognition with Low Level Descriptor Features

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

Fusion Model for Speech Emotion Recognition with Low Level D...

第十四届全国人机语音通讯学术会议

作者： Cheng Chang Huifeng Zhang Zhangxuan Gu Yanmin Qian key laboratory of shanghai education commission for intelligent interaction and cognitive engineering SpeechLab Department of Computer Science and Engineering Brain Science and Technology Research Center Shanghai Jiao Tong University

Speech emotion recognition is one of the most challenging speech processing tasks,with many applications in the field of Human-Machine interaction（HMI）.Traditional works have been using Gaussian Mixture Models（GMM） for classification.In recent researches,Deep Neural Networks（DNN） have shown strong ability in feature learning and modeling in many tasks.In this paper,we present to utilize the DNN to learn extra features from Low Level Descriptor（LLD） features for other classifiers,and propose a decision-level fusion model for speech emotion recognition.We carry out experiments and evaluations on a public German corpus called FAU-AIBO with five emotions.And our experimental results using LLD features demonstrate that our proposed approach improves the recognition performance on unweighted results and outperforms the baseline using GMM significantly.

关键词： speech emotion recognition deep neural network low level descriptor fusion

Speech Emotion Recognition Based on SVM and GMM-HMM Hybrid System

在线全文

cnki会议

学校读者我要写书评

暂无评论

Speech Emotion Recognition Based on SVM and GMM-HMM Hybrid S...

第十四届全国人机语音通讯学术会议

作者： Kaiyu Shi Xuan Liu Yanmin Qian key laboratory of shanghai education commission for intelligent interaction and cognitive engineering SpeechLab Department of Computer Science and Engineering Brain Science and Technology Research Center Shanghai Jiao Tong University

Speech emotion recognition is one of the latest challenges in speech processing,we implemented a hybrid system for speech emotion recognition.Two methods were proposed,compared and combined.At first,we utilized GMM-HMM model fed with MFCC feature to exploit the dynamics of emotional signals.Then a balanced SVM classifier was applied on the static LLD feature.In order to take advantage of generative model and discriminate model,these two methods were combined by taking the probability representation of the GMM-HMM model as the additional features for the SVM classifier.This hybrid system makes full use of both MFCC feature and LLD feature,the performance of the hybrid system also proves the effectiveness of our proposed system.

关键词： speech emotion recognition SVM GMMHMM hybrid system

Emotion Recognition Using Support Vector Machine and Deep Neural Network

在线全文

cnki会议

学校读者我要写书评

暂无评论

Emotion Recognition Using Support Vector Machine and Deep Ne...

第十四届全国人机语音通讯学术会议

作者： Chen Ruinian Zhou Ying Qian Yanmin key laboratory of shanghai education commission for intelligent interaction and cognitive engineering SpeechLab Department of Computer Science and Engineering Brain Science and Technology Research Center Shanghai Jiao Tong University

We propose a classification system which is a combination of three subsystem to solve the speech emotion recognition problem.The first subsystem utilizes support vector machines(SVMs) to classify the features directly.The second subsystem utilizes deep neural networks(DNNs) to classify the features directly.In the third subsystem,we utilize DNNs to extract high level features from raw data and show that they are effective for speech emotion recognition.We first produce an emotion state probability distribution for each speech segment using DNNs,then we construct utterance-level features from segment-level probability distributions.These utterance level features are fed into an SVM to identify utterance-level emotions.The experimental results show that all the subsystems are outperform the HMM baseline system,and the combined system get the best result on F-score.

关键词： Emotion recognition Deep neural networks Support vector machine

GLEU-guided Multi-Resolution Network for Short Text Conversation

在线全文

cnki会议

学校读者我要写书评

暂无评论

GLEU-guided Multi-Resolution Network for Short Text Conversa...

第十四届全国人机语音通讯学术会议

作者： Xuan Liu Kai Yu key laboratory of shanghai education commission for intelligent interaction and cognitive engineering SpeechLab Department of Computer Science and Engineering Brain Science and Technology Research Center Shanghai Jiao Tong University

With the recent development of sequence-to-sequence framework,generation approach for short text conversation becomes attractive.Traditional sequence-to-sequence method for short text conversation often suffers from dull response problem.Multi-resolution generation approach has been introduced to address this problem by dividing the generation process into two steps:keywords-sequence generation and response generation.However,this method still tends to generate short and common keywords-sequence.In this work,a new multi-resolution generation framework is proposed.Instead of using the word-level maximum likelihood criterion,we optimize the sequence-level GLEU score of the entire generated keywords-sequence using a policy gradient approach in reinforcement learning.Experiments show that the proposed approach can generate longer and more diverse keywords-sequence.Meanwhile,it achieves better score in human evaluation.

关键词： short text conversation sequence-to-sequence policy gradient