Filters








30,108 Hits in 1.1 sec

Streaming Chunk-Aware Multihead Attention for Online End-to-End Speech Recognition [article]

Shiliang Zhang, Zhifu Gao, Haoneng Luo, Ming Lei, Jie Gao, Zhijie Yan, Lei Xie
2020 arXiv   pre-print
Recently, streaming end-to-end automatic speech recognition (E2E-ASR) has gained more and more attention. Many efforts have been paid to turn the non-streaming attention-based E2E-ASR system into streaming architecture. In this work, we propose a novel online E2E-ASR system by using Streaming Chunk-Aware Multihead Attention(SCAMA) and a latency control memory equipped self-attention network (LC-SAN-M). LC-SAN-M uses chunk-level input to control the latency of encoder. As to SCAMA, a jointly
more » ... ned predictor is used to control the output of encoder when feeding to decoder, which enables decoder to generate output in streaming manner. Experimental results on the open 170-hour AISHELL-1 and an industrial-level 20000-hour Mandarin speech recognition tasks show that our approach can significantly outperform the MoChA-based baseline system under comparable setup. On the AISHELL-1 task, our proposed method achieves a character error rate (CER) of 7.39%, to the best of our knowledge, which is the best published performance for online ASR.
arXiv:2006.01712v1 fatcat:6fpcj2y2bfgf5c5xkdewioytc4

Comprehensive routing strategy on multilayer networks [article]

Lei Gao, Panpan Shu, Ming Tang, Wei Wang, Hui Gao
2017 arXiv   pre-print
Designing an efficient routing strategy is of great importance to alleviate traffic congestion in multilayer networks. In this work, we design an effective routing strategy for multilayer networks by comprehensively considering the roles of nodes' local structures in micro-level, as well as the macro-level differences in transmission speeds between different layers. Both numerical and analytical results indicate that our proposed routing strategy can reasonably redistribute the traffic load of
more » ... he low speed layer to the high speed layer, and thus the traffic capacity of multilayer networks are significantly enhanced compared with the monolayer low speed networks. There is an optimal combination of macro- and micro-level control parameters at which can remarkably alleviate the congestion and thus maximize the traffic capacity for a given multilayer network. Moreover, we find that increasing the size and the average degree of the high speed layer can enhance the traffic capacity of multilayer networks more effectively. We finally verify that real-world network topology does not invalidate the results. The theoretical predictions agree well with the numerical simulations.
arXiv:1701.05307v1 fatcat:f7lqsy6bsrf4jd4jvsgdsu43pm

The application of co-integration theory in ensemble pulsar timescale algorithm [article]

Feng Gao, Ming-Lei Tong, Yu-Ping Gao, Ting-Gao Yang, Cheng-Shi Zhao
2019 arXiv   pre-print
We can define the dispersion of residual as σ RMS , while white noise is denoted σ W and red noise is denoted σ TN (Yang et al. 2014; Gao et al. 2018) .  ...  (Gao et al. 2018; Lam et al. 2017) where t stands for timing span, V ar(X(t)) represents the variance of the data in t, and τ is the increment of timing span.  ... 
arXiv:1902.07072v1 fatcat:4ubkrabncjbw5bxigx4loux2k4

Extended Batch Normalization [article]

Chunjie Luo, Jianfeng Zhan, Lei Wang, Wanling Gao
2020 arXiv   pre-print
Batch normalization (BN) has become a standard technique for training the modern deep networks. However, its effectiveness diminishes when the batch size becomes smaller, since the batch statistics estimation becomes inaccurate. That hinders batch normalization's usage for 1) training larger model which requires small batches constrained by memory consumption, 2) training on mobile or embedded devices of which the memory resource is limited. In this paper, we propose a simple but effective
more » ... d, called extended batch normalization (EBN). For NCHW format feature maps, extended batch normalization computes the mean along the (N, H, W) dimensions, as the same as batch normalization, to maintain the advantage of batch normalization. To alleviate the problem caused by small batch size, extended batch normalization computes the standard deviation along the (N, C, H, W) dimensions, thus enlarges the number of samples from which the standard deviation is computed. We compare extended batch normalization with batch normalization and group normalization on the datasets of MNIST, CIFAR-10/100, STL-10, and ImageNet, respectively. The experiments show that extended batch normalization alleviates the problem of batch normalization with small batch size while achieving close performances to batch normalization with large batch size.
arXiv:2003.05569v1 fatcat:stjet3y3ubbnlji2xbep36pb6m

Zerumbone inhibits migration in ESCC via promoting Rac1 ubiquitination

Meilin Wang, Jingling Niu, Lei Gao, Yang Gao, Shegan Gao
2019 Biomedicine and Pharmacotherapy  
Authors' contributions Supervision of study: Shegan Gao. Conflicts of interest The authors declare that they have no conflicts of interest concerning this article.  ... 
doi:10.1016/j.biopha.2018.11.134 fatcat:4psbqjlaobhlxldusq5cisuuzi

Pion production and absorption in heavy-ion collisions [article]

Yuan Gao, Lei Zhang, Gao-Chan Yong, Zi-Yu Liu, Wei Zuo
2020 arXiv   pre-print
Based on the isospin dependent transport model IBUU, the pion production and its absorption are thoroughly studied in the central collision of Au+Au at a beam energy of 400 MeV/nucleon. It is found that the pions are firstly produced by the hard Δ decay at the average density around 1.75ρ_0, whereas about 18% of them are absorbed absolutely in the subsequent inelastic collisions. For the free pions observed, more than half of them have been scattered for one or more times before they are free
more » ... om matter. And the more scattering numbers of pions, the higher the momentum they possess. These pions, due to longer time of their existence in high density nuclear matter, carry more information about the symmetry energy of the nuclear matter at high densities.
arXiv:2009.02674v1 fatcat:4ivymnzdyvfd3cl64f2trgw6fm

Risk and prognostic factors of breast cancer with liver metastases

Lei Ji, Lei Cheng, Xiuzhi Zhu, Yu Gao, Lei Fan, Zhonghua Wang
2021 BMC Cancer  
Acknowledgements We acknowledge Xiuzhi Zhu and Yu Gao for collecting data, Zhangru yang and Jianjiao Ni for critically reviewing the manuscript, and Lei Fan and ZhongHua Wang for their leadership and guidance  ... 
doi:10.1186/s12885-021-07968-5 pmid:33676449 fatcat:cefgppoj4jekjepkcxyziyugpa

On thurston's core entropy algorithm dedicated to the memory of tan lei [article]

Yan Gao
2017 arXiv   pre-print
Gao-L. Tan ( [21] ) and W. Jung ([10] ).  ...  The initial purpose, as suggested by my supervisor Tan Lei, was to fill in the empty section "Hausdorff dimension and growth rate" of W.  ... 
arXiv:1511.06513v2 fatcat:rmnhg5vdlbebhdxwt5chffce7m

Streaming Chunk-Aware Multihead Attention for Online End-to-End Speech Recognition

Shiliang Zhang, Zhifu Gao, Haoneng Luo, Ming Lei, Jie Gao, Zhijie Yan, Lei Xie
2020 Interspeech 2020  
Recently, streaming end-to-end automatic speech recognition (E2E-ASR) has gained more and more attention. Many efforts have been paid to turn the non-streaming attention-based E2E-ASR system into streaming architecture. In this work, we propose a novel online E2E-ASR system by using Streaming Chunk-Aware Multihead Attention (SCAMA) and a latency control memory equipped self-attention network (LC-SAN-M). LC-SAN-M uses chunk-level input to control the latency of encoder. As to SCAMA, a jointly
more » ... ined predictor is used to control the output of encoder when feeding to decoder, which enables decoder to generate output in streaming manner. Experimental results on the open 170-hour AISHELL-1 and an industrial-level 20000-hour Mandarin speech recognition tasks show that our approach can significantly outperform the MoChA-based baseline system under comparable setup. On the AISHELL-1 task, our proposed method achieves a character error rate (CER) of 7.39%, to the best of our knowledge, which is the best published performance for online ASR.
doi:10.21437/interspeech.2020-1972 dblp:conf/interspeech/ZhangGLLGYX20 fatcat:qes5faccuvcobkbrgkhl4xrcea

Differentiable Meta-learning Model for Few-shot Semantic Segmentation [article]

Pinzhuo Tian, Zhangkai Wu, Lei Qi, Lei Wang, Yinghuan Shi, Yang Gao
2019 arXiv   pre-print
Acknowledge The work of Yang Gao is supported by Science and Technology Innovation 2030-"New Generation Artificial Intelligence" Major Project (No. 2018AAA0100905), NSFC (No. 61432008) and the Collaborative  ... 
arXiv:1911.10371v1 fatcat:5tu4krpmlvbmlezthin3i2tqna

TUNABLE LATERAL SHIFT THROUGH NONLINEAR COMPOSITES OF NONSPHERICAL PARTICLES

Dongliang Gao, Lei Gao
2009 Electromagnetic Waves  
The Goos-Hänchen (GH) shifts of the reflected waves from nonlinear nanocomposites of interleaved nonspherical metal and dielectric particles are investigated both theoretically and numerically. First, based on spectral representation theory and effective medium approximation, we derive the field-dependent effective permittivity of nonlinear composites. Then, the stationary phase method is adopted to study the GH shifts from nonlinear composites. It is found that for a given volume fraction,
more » ... e exist two critical polarization factors L c1 and L c2 , and bistable GH shifts appear only when L < L c1 or L > L c2 . Moreover, both giant negative and positive GH shifts accompanied with large reflectivity are found, hence they can be easily observed in experiments. The reversal of the GH shift may be controlled by adjusting both the incident angle and the applied field. Numerical simulations for Gaussian-type incident beam are performed, and good agreement between simulated data and theoretical ones is found especially for large waist width.
doi:10.2528/pier09102404 fatcat:yo4dsdunabhmhex6o2ss3fjka4

Genetic correlation between current circulating H1N1 swine and human influenza viruses

Lu Lu, Yanbo Yin, Zhongsheng Sun, Lei Gao, George F. Gao, Sidang Liu, Lei Sun, Wenjun Liu
2010 Journal of Clinical Virology  
Background: H1N1 is the main subtype influenza A virus circulating in human and swine population, and has long been a threat to economy and public health. Objective: To explore the genetic correlation between current circulating H1N1 swine and human influenza viruses. Study design: Three new H1N1 swine influenza viruses (SIVs) were isolated and genomes sequencing were conducted followed by phylogenetic and molecular analysis of all swine and human H1N1 influenza viruses isolated in China in the past five years.
doi:10.1016/j.jcv.2010.07.018 pmid:20732830 fatcat:7hmqfgze6zbo3gshdttg5zseki

Efficient Generation of Mice with Consistent Transgene Expression by FEEST

Lei Gao, Yonghua Jiang, Libing Mu, Yanbin Liu, Fengchao Wang, Peng Wang, Aiqun Zhang, Nan Tang, Ting Chen, Minmin Luo, Lei Yu, Shaorong Gao (+1 others)
2015 Scientific Reports  
doi:10.1038/srep16284 pmid:26573149 pmcid:PMC4648098 fatcat:pjmizc26ifbgvlxuwbcd36gkou

Convergent evidence from systematic analysis of GWAS revealed genetic basis of esophageal cancer

Xue-xin Gao, Lei Gao, Jiu-qiang Wang, Su-su Qu, Yue Qu, Hong-lei Sun, Si-dang Liu, Ying-li Shang
2016 OncoTarget  
Recent genome-wide association studies (GWAS) have identified single nucleotide polymorphisms (SNPs) associated with risk of esophageal cancer (EC). However, investigation of genetic basis from the perspective of systematic biology and integrative genomics remains scarce. In this study, we explored genetic basis of EC based on GWAS data and implemented a series of bioinformatics methods including functional annotation, expression quantitative trait loci (eQTL) analysis, pathway enrichment
more » ... is and pathway grouped network analysis. Two hundred and thirteen risk SNPs were identified, in which 44 SNPs were found to have significantly differential gene expression in esophageal tissues by eQTL analysis. By pathway enrichment analysis, 170 risk genes mapped by risk SNPs were enriched into 38 significant GO terms and 17 significant KEGG pathways, which were significantly grouped into 9 sub-networks by pathway grouped network analysis. The 9 groups of interconnected pathways were mainly involved with muscle cell proliferation, cellular response to interleukin-6, cell adhesion molecules, and ethanol oxidation, which might participate in the development of EC. Our findings provide genetic evidence and new insight for exploring the molecular mechanisms of EC.
doi:10.18632/oncotarget.10133 pmid:27331408 pmcid:PMC5190123 fatcat:nmzox6bvlzdwnmfjtpo4xbhwve

Gauge invariance in phase-gradient metagratings [article]

Qingjia zhou, Lei Gao, Yadong Xu
2021 arXiv   pre-print
Phase gradient metagratings/metasurfaces (PGMs) have provided a new paradigm for light manipulations. In this work, we will show the existence of gauge invariance in PGMs, i.e., the diffraction law of PGMs is independent of the choice of initial value of abrupt phase shift that induces the phase gradient. This gauge invariance ensures the well-studied ordinary metallic grating that can be regarded as a PGM, with its diffraction properties that can fully predicted by generalized diffraction law
more » ... ith phase gradient. The generalized diffraction law presents a new insight for the famous effect of Wood's Anomalies and Rayleigh conjecture.
arXiv:2107.01868v1 fatcat:axirnqc2yjei5bxq2t2dmmolbm
« Previous Showing results 1 — 15 out of 30,108 results