arXiv subjects

Ke Hu

Publications and source records attributed to Ke Hu.

At least 73 records · Page 4Linked to original sources

arXiv2020

Multimillijoule terahertz radiation from laser interactions with microplasma-waveguides

When a relativistic, femtosecond laser pulse enters a waveguide, the pulse energy is coupled into waveguide optical modes. The longitudinal laser field effectively accelerates electrons along the axis of the channel, while the asymmetric transverse electromagnetic fields tend to expel fast electrons radially outwards. At the exit of the waveguide, the $\sim$${\rm nC}$, $\sim$$10\ {\rm MeV}$ electron bunch converts its energy to a $\sim$$10\ {\rm mJ}$ terahertz (THz) laser pulse through coherent diffraction radiation. In this paper, we present 3D particle-in-cell simulations and theoretical analyses of the aforementioned interaction process. We investigate the process of longitudinal acceleration and radial expulsion of fast electrons, as well as the dependence of the properties of the resulting THz radiation on laser and plasma parameters and the effects of the preplasma. The simulation results indicate that the conversion efficiency of energy can be over $5\%$ if the waveguide length is optimal and a high contrast pump laser is used. These results guide the design of more intense and powerful THz sources.

physics.plasm-ph↗

arXiv2020

HazeDose: Design and Analysis of a Personal Air Pollution Inhaled Dose Estimation System using Wearable Sensors

Nowadays air pollution becomes one of the biggest world issues in both developing and developed countries. Helping individuals understand their air pollution exposure and health risks, the traditional way is to utilize data from static monitoring stations and estimate air pollution qualities in a large area by government agencies. Data from such sensing system is very sparse and cannot reflect real personal exposure. In recent years, several research groups have developed participatory air pollution sensing systems which use wearable or portable units coupled with smartphones to crowd-source urban air pollution data. These systems have shown remarkable improvement in spatial granularity over government-operated fixed monitoring systems. In this paper, we extend the paradigm to HazeDose system, which can personalize the individuals' air pollution exposure. Specifically, we combine the pollution concentrations obtained from an air pollution estimation system with the activity data from the individual's on-body activity monitors to estimate the personal inhalation dosage of air pollution. Users can visualize their personalized air pollution exposure information via a mobile application. We show that different activities, such as walking, cycling, or driving, impact their dosage, and commuting patterns contribute to a significant proportion of an individual's daily air pollution dosage. Moreover, we propose a dosage minimization algorithm, with the trial results showing that up to 14.1% of a biker's daily exposure can be reduced while using alternative routes the driver can inhale 25.9% less than usual. One heuristic algorithm is also introduced to balance the execution time and dosage reduction for alternative routes scenarios. The results show that up to 20.3% dosage reduction can be achieved when the execution time is almost one seventieth of the original one.

cs.CY↗

arXiv2020

A Streaming On-Device End-to-End Model Surpassing Server-Side Conventional Model Quality and Latency

Thus far, end-to-end (E2E) models have not been shown to outperform state-of-the-art conventional models with respect to both quality, i.e., word error rate (WER), and latency, i.e., the time the hypothesis is finalized after the user stops speaking. In this paper, we develop a first-pass Recurrent Neural Network Transducer (RNN-T) model and a second-pass Listen, Attend, Spell (LAS) rescorer that surpasses a conventional model in both quality and latency. On the quality side, we incorporate a large number of utterances across varied domains to increase acoustic diversity and the vocabulary seen by the model. We also train with accented English speech to make the model more robust to different pronunciations. In addition, given the increased amount of training data, we explore a varied learning rate schedule. On the latency front, we explore using the end-of-sentence decision emitted by the RNN-T model to close the microphone, and also introduce various optimizations to improve the speed of LAS rescoring. Overall, we find that RNN-T+LAS offers a better WER and latency tradeoff compared to a conventional model. For example, for the same latency, RNN-T+LAS obtains a 8% relative improvement in WER, while being more than 400-times smaller in model size.

cs.CL↗

arXiv2020

Studying the local magnetic field and anisotropy of magnetic turbulence by synchrotron polarization derivative

Due to the inevitable accumulation of the observed information in the direction of the line of sight, it is difficult to measure the local magnetic field of MHD turbulence. However, the correct understanding of the local magnetic field is a prerequisite for reconstructing the Galactic 3D magnetic field. We study how to reveal the local magnetic field direction and the eddy anisotropy on the basis of the statistics of synchrotron polarization derivative with respect to the squared wavelength $dP/dλ^2$. In the low frequency and strong Faraday rotation regime, we implement numerical simulations in the combination of multiple statistic techniques, such as structure function, quadrupole ratio modulus, spectral correlation function, correlation function anisotropy and spatial gradient techniques. We find that (1) statistic analysis of $dP/dλ^2$ indeed reveals the anisotropy of underlying MHD turbulence, the degree of which increases with the increase of the radiation frequency; (2) the synergy of both correlation function anisotropy and gradient calculation of $dP/dλ^2$ enables the measurement of the local magnetic field direction.

astro-ph.HE↗

arXiv2020

Highly-efficient terahertz radiation generated by surface electrons from laser-foil interactions

A novel scheme for generating powerful terahertz (THz) radiation based on laser-solid interactions is proposed. When a $p$-polarized femtosecond laser impinges obliquely on a plane solid target and the target partially blocks the laser energy, surface electrons are extracted out and accelerated by the laser fields, forming a low-divergence electron beam. A half-cycle THz radiation pulse is emitted simultaneously as the beam passes by the edge of the target, due to coherent diffraction radiation. Our particle-in-cell simulations show that the relativistic THz pulse can have an energy of a few tens of millijoule and the conversion efficiency can be over 1$\%$ with existing $\sim$J level femtosecond laser sources.

physics.plasm-ph↗

arXiv2020

Deliberation Model Based Two-Pass End-to-End Speech Recognition

End-to-end (E2E) models have made rapid progress in automatic speech recognition (ASR) and perform competitively relative to conventional models. To further improve the quality, a two-pass model has been proposed to rescore streamed hypotheses using the non-streaming Listen, Attend and Spell (LAS) model while maintaining a reasonable latency. The model attends to acoustics to rescore hypotheses, as opposed to a class of neural correction models that use only first-pass text hypotheses. In this work, we propose to attend to both acoustics and first-pass hypotheses using a deliberation network. A bidirectional encoder is used to extract context information from first-pass hypotheses. The proposed deliberation model achieves 12% relative WER reduction compared to LAS rescoring in Google Voice Search (VS) tasks, and 23% reduction on a proper noun test set. Compared to a large conventional model, our best model performs 21% relatively better for VS. In terms of computational complexity, the deliberation decoder has a larger size than the LAS decoder, and hence requires more computations in second-pass decoding.

eess.AS↗

arXiv2019

Spin-orbit interaction in a high-power laser irradiated micro-scale plasma waveguide

Light carries angular momentum as spin and orbital components. The spin-orbit interaction (SOI) of light refers to phenomena in which the spin (left or right circular polarisation) affects the spatial degrees of freedom. Recently, interest in SOI has surged, as it provides physical insight into the behaviour of polarised light at subwavelength scales, and allows for spin-controlled manipulation of light. Most studies are performed with low intensity, leaving the role SOI plays in the relativistic laser-plasma interaction, characterised by nonlinearity, less understood. Here, using 3D particle-in-cell simulations, we report SOI effects in this unprecedented regime. Specifically, a circularly polarised Gaussian laser irradiating a micro-scale plasma waveguide drives a spin-controlled chiral surface wave, allowing the reflected harmonic photons to gain orbital angular momentum (OAM). These effects produce intense optical vortices in the extreme ultraviolet regime - an area of significant fundamental and applied physics potential.

physics.plasm-ph↗

arXiv2019

Phoneme-Based Contextualization for Cross-Lingual Speech Recognition in End-to-End Models

Contextual automatic speech recognition, i.e., biasing recognition towards a given context (e.g. user's playlists, or contacts), is challenging in end-to-end (E2E) models. Such models maintain a limited number of candidates during beam-search decoding, and have been found to recognize rare named entities poorly. The problem is exacerbated when biasing towards proper nouns in foreign languages, e.g., geographic location names, which are virtually unseen in training and are thus out-of-vocabulary (OOV). While grapheme or wordpiece E2E models might have a difficult time spelling OOV words, phonemes are more acoustically salient and past work has shown that E2E phoneme models can better predict such words. In this work, we propose an E2E model containing both English wordpieces and phonemes in the modeling space, and perform contextual biasing of foreign words at the phoneme level by mapping pronunciations of foreign words into similar English phonemes. In experimental evaluations, we find that the proposed approach performs 16% better than a grapheme-only biasing model, and 8% better than a wordpiece-only biasing model on a foreign place name recognition task, with only slight degradation on regular English tasks.

cs.CL↗

arXiv2019

Adversarial Training for Multilingual Acoustic Modeling

Multilingual training has been shown to improve acoustic modeling performance by sharing and transferring knowledge in modeling different languages. Knowledge sharing is usually achieved by using common lower-level layers for different languages in a deep neural network. Recently, the domain adversarial network was proposed to reduce domain mismatch of training data and learn domain-invariant features. It is thus worth exploring whether adversarial training can further promote knowledge sharing in multilingual models. In this work, we apply the domain adversarial network to encourage the shared layers of a multilingual model to learn language-invariant features. Bidirectional Long Short-Term Memory (LSTM) recurrent neural networks (RNN) are used as building blocks. We show that shared layers learned this way contain less language identification information and lead to better performance. In an automatic speech recognition task for seven languages, the resultant acoustic model improves the word error rate (WER) of the multilingual model by 4% relative on average, and the monolingual models by 10%.

cs.CL↗

arXiv2018

Attaining the Unattainable? Reassessing Claims of Human Parity in Neural Machine Translation

We reassess a recent study (Hassan et al., 2018) that claimed that machine translation (MT) has reached human parity for the translation of news from Chinese into English, using pairwise ranking and considering three variables that were not taken into account in that previous study: the language in which the source side of the test set was originally written, the translation proficiency of the evaluators, and the provision of inter-sentential context. If we consider only original source text (i.e. not translated from another language, or translationese), then we find evidence showing that human parity has not been achieved. We compare the judgments of professional translators against those of non-experts and discover that those of the experts result in higher inter-annotator agreement and better discrimination between human and machine translations. In addition, we analyse the human translations of the test set and identify important translation issues. Finally, based on these findings, we provide a set of recommendations for future human evaluations of MT.

cs.CL↗

arXiv2017

Predicting disease-related genes by path-based similarity and community structure in protein-protein interaction network

Network-based computational approaches to predict unknown genes associated with certain diseases are of considerable significance for uncovering the molecular basis of human diseases. In this paper, we proposed a kind of new disease-gene-prediction methods by combining the path-based similarity with the community structure in the human protein-protein interaction network. Firstly, we introduced a set of path-based similarity indices, a novel community-based similarity index, and a new similarity combining the path-based similarity index. Then we assessed the statistical significance of the measures in distinguishing the disease genes from non-disease genes, to confirm their availability in predicting disease genes. Finally, we applied these measures to the disease-gene prediction of single disease-gene family, and analyzed the performance of these measures in disease-gene prediction, especially the effect of the community structure on the prediction performance in detail. The results indicated that genes associated with the same or similar diseases commonly reside in the same community of the protein-protein interaction network, and the community structure is greatly helpful for the disease-gene prediction.

q-bio.MN↗

arXiv2016

Enhancing community detection by local structural information

Many real-world networks such as the gene networks, protein-protein interaction networks and metabolic networks exhibit community structures, meaning the existence of groups of densely connected vertices in the networks. Many local similarity measures in the networks are closely related to the concept of the community structures, and may have positive effect on community detection in the networks. Here, various local similarity measures are used to extract the local structural information and then are applied to community detection in the networks by using the edge-reweighting strategy. The effect of the local similarity measures on community detection is carefully investigated and compared in various networks. The experimental results show that the local similarity measures are crucial to the improvement for the community detection methods, while the positive effect of the local similarity measures is closely related to the networks under study and the applied community detection methods.

physics.soc-ph↗

arXiv2015

Analysis and perturbation of degree correlation in complex networks

Degree correlation is an important topological property common to many real-world networks. In this paper, the statistical measures for characterizing the degree correlation in networks are investigated analytically. We give an exact proof of the consistency for the statistical measures, reveal the general linear relation in the degree correlation, which provide a simple and interesting perspective on the analysis of the degree correlation in complex networks. By using the general linear analysis, we investigate the perturbation of the degree correlation in complex networks caused by the addition of few nodes and the rich club. The results show that the assortativity of homogeneous networks such as the ER graphs is easily to be affected strongly by the simple structural changes, while it has only slight variation for heterogeneous networks with broad degree distribution such as the scale-free networks. Clearly, the homogeneous networks are more sensitive for the perturbation than the heterogeneous networks.

physics.soc-ph↗

arXiv2015

Multi-resolution community detection based on generalized self-loop rescaling strategy

Community detection is of considerable importance for analyzing the structure and function of complex networks. Many real-world networks may possess community structures at multiple scales, and recently, various multi-resolution methods were proposed to identify the community structures at different scales. In this paper, we present a type of multi-resolution methods by using the generalized self-loop rescaling strategy. The self-loop rescaling strategy provides one uniform ansatz for the design of multi-resolution community detection methods. Many quality functions for community detection can be unified in the framework of the self-loop rescaling. The resulting multi-resolution quality functions can be optimized directly using the existing modularity-optimization algorithms. Several derived multi-resolution methods are applied to the analysis of community structures in several synthetic and real-world networks. The results show that these methods can find the pre-defined substructures in synthetic networks and real splits observed in real-world networks. Finally, we give a discussion on the methods themselves and their relationship. We hope that the study in the paper can be helpful for the understanding of the multi-resolution methods and provide useful insight into designing new community detection methods.

physics.soc-ph↗

arXiv2013

Robustness of Link-prediction Algorithm Based on Similarity and Application to Biological Networks

Many algorithms have been proposed to predict missing links in a variety of real networks. These studies focus on mainly both accuracy and efficiency of these algorithms. However, little attention is paid to their robustness against either noise or irrationality of a link existing in almost all of real networks. In this paper, we investigate the robustness of several typical node-similarity-based algorithms and find that these algorithms are sensitive to the strength of noise. Moreover, we find that it also depends on networks' structure properties, especially on network efficiency, clustering coefficient and average degree. In addition, we make an attempt to enhance the robustness by using link weighting method to transform un-weighted network to weighted one and then make use of weights of links to characterize their reliability. The result shows that proper link weighting scheme can enhance both robustness and accuracy of these algorithms significantly in biological networks while it brings little computational effort.

physics.soc-ph↗

arXiv2012

Link Prediction in Complex Networks by Multi Degree Preferential-Attachment Indices

In principle, the rules of links formation of a network model can be considered as a kind of link prediction algorithm. By revisiting the preferential attachment mechanism for generating a scale-free network, here we propose a class of preferential attachment indices which are different from the previous one. Traditionally, the preferential attachment index is defined by the product of the related nodes degrees, while the new indices will define the similarity score of a pair of nodes by either the maximum in the two nodes degrees or the summarization of their degrees. Extensive experiments are carried out on fourteen real-world networks. Compared with the traditional preferential attachment index, the new ones, especially the degree-summarization similarity index, can provide more accurate prediction on most of the networks. Due to the improved prediction accuracy and low computational complexity, these proposed preferential attachment indices may be of help to provide an instruction for mining unknown links in incomplete networks.

physics.soc-ph↗

arXiv2011

Immunization for complex network based on the effective degree of vertex

The basic idea of many effective immunization strategies is first to rank the importance of vertices according to the degrees of vertices and then remove the vertices from highest importance to lowest until the network becomes disconnected. Here we define the effective degrees of vertex, i.e., the number of its connections linking to un-immunized nodes in current network during the immunization procedure, to rank the importance of vertex, and modify these strategies by using the effective degrees of vertices. Simulations on both the scale-free network models with various degree correlations and two real networks have revealed that the immunization strategies based on the effective degrees are often more effective than those based on the degrees in the initial network.

physics.soc-ph↗

arXiv2011

Limitation of multi-resolution methods in community detection

Recently, a type of multi-resolution methods in community detection was introduced, which can adjust the resolution of modularity by modifying the modularity function with tunable resolution parameters, such as those proposed by Arenas, Fernandez and Gomez and by Reichardt and Bornholdt. In this paper, we show that these methods still have the intrinsic limitation-large communities may have been split before small communities become visible-because it is at the cost of the community stability that the enhancement of the modularity resolution is obtained. The theoretical results indicated that the limitation depends on the degree of interconnectedness of small communities and the difference between the sizes of small communities and of large communities, while independent of the size of the whole network. These findings have been confirmed in several example networks, where communities even are full-completed sub-graphs.

physics.soc-ph↗