Language model cross adaptation for LVCSR system combination

State-of-the-art large vocabulary continuous speech recognition (LVCSR) systems often combine outputs from multiple sub-systems that may even be developed at different sites. Cross system adaptation, in which model adaptation is performed using the outputs from another sub-system, can be used as an alternative to hypothesis level combination schemes such as ROVER. Normally cross adaptation is only performed on the acoustic models. However, there are many other levels in LVCSR systems' modelling hierarchy where complimentary features may be exploited, for example, the sub-word and the word level, to further improve cross adaptation based system combination. It is thus interesting to also cross adapt language models (LMs) to capture these additional useful features. In this paper cross adaptation is applied to three forms of language models, a multi-level LM that models both syllable and word sequences, a word level neural network LM, and the linear combination of the two. Significant error rate reductions of 4.0-7.1% relative were obtained over ROVER and acoustic model only cross adaptation when combining a range of Chinese LVCSR sub-systems used in the 2010 and 2011 DARPA GALE evaluations. © 2012 Elsevier Ltd. All rights reserved.

Paraphrastic language models

In natural languages multiple word sequences can represent the same underlying meaning. Only modelling the observed surface word sequence can result in poor context coverage, for example, when using n-gram language models (LM). To handle this issue, this paper presents a novel form of language model, the paraphrastic LM. A phrase level transduction model that is statistically learned from standard text data is used to generate paraphrase variants. LM probabilities are then estimated by maximizing their marginal probability. Significant error rate reductions of 0.5%-0.6% absolute were obtained on a state-ofthe-art conversational telephone speech recognition task using a paraphrastic multi-level LM modelling both word and phrase sequences.

Paraphrastic language models and combination with neural network language models

In natural languages multiple word sequences can represent the same underlying meaning. Only modelling the observed surface word sequence can result in poor context coverage, for example, when using n-gram language models (LM). To handle this issue, paraphrastic LMs were proposed in previous research and successfully applied to a US English conversational telephone speech transcription task. In order to exploit the complementary characteristics of paraphrastic LMs and neural network LMs (NNLM), the combination between the two is investigated in this paper. To investigate paraphrastic LMs' generalization ability to other languages, experiments are conducted on a Mandarin Chinese broadcast speech transcription task. Using a paraphrastic multi-level LM modelling both word and phrase sequences, significant error rate reductions of 0.9% absolute (9% relative) and 0.5% absolute (5% relative) were obtained over the baseline n-gram and NNLM systems respectively, after a combination with word and phrase level NNLMs. © 2013 IEEE.

Single-photon source in strong photon-atom interaction regime in quasiperiodic photonic crystals

The antibunching properties of the fluorescence from a two-level ideal system in a 12-fold quasiperiodic photonic crystal are investigated based on the calculated local density of states. We found that the antibunching phenomenon of the fluorescence from two-level ideal systems could be significantly changed by varying their positions, i.e., perfect antibunching and antibunching with damped Rabi oscillation phenomenon occurred in different positions and at different frequencies in photonic crystals as a result of the large differences in the local density of states. This study revealed that the multi-level coherence of fluorescence from a two-level ideal system could be manipulated by controlling the position of the two-level ideal system in photonic crystals and the emission frequency in the photonic band structure. Copyright (C) EPLA, 2008

多级多版本数据库管理系统全局串行化

多级调度应该保证事务历史可串行化,满足多级安全特性,不会引入隐通道,并保证高级别事务不会因为无限等待而"饿死".与其他多级数据管理系统调度机制相比,多级多版本时戳调度机制满足上述要求,但该机制存在两个问题,一是事务可能读旧版本,二是要求调度器是可信进程.提出一种多级多版本全局时戳调度机制(MLS_MVGTO),以及依据事务快照生成其全局时戳的基本步骤.给出了预知只读事务信息时的两种改进方法.MLS_MVGTO机制生成的事务历史可串行化,不引入隐通道等,并且该方法避免引入一个全局可信的调度器,并通过对只读事务的深入分析,允许事务读新版本.

一种实用的安全数据库多级关系模型

安全数据库管理系统的安全策略可以为数据库提供强大的安全功能,现在BLP模型是用得比较普遍的策略模型之一。为了适应BLP模型,采用一种实用的多级关系模型作为安全数据库的基础,这是原有关系模型的扩展。该文描述了多级关系模型的原理和内容,特别对多级关系模型的分解与合并做了详细的阐述。并说明了这个模型适合于在多级安全操作系统上实现多级数据库管理系统。

Smoothing LDA Model for Text Categorization

Abstract. Latent Dirichlet Allocation (LDA) is a document level language model. In general, LDA employ the symmetry Dirichlet distribution as prior of the topic-words’ distributions to implement model smoothing. In this paper, we propose a data-driven smoothing strategy in which probability mass is allocated from smoothing-data to latent variables by the intrinsic inference procedure of LDA. In such a way, the arbitrariness of choosing latent variables'priors for the multi-level graphical model is overcome. Following this data-driven strategy,two concrete methods, Laplacian smoothing and Jelinek-Mercer smoothing, are employed to LDA model. Evaluations on different text categorization collections show data-driven smoothing can significantly improve the performance in balanced and unbalanced corpora.

一个支持可信主体特权最小化的多级安全模型

对已有多级安全模型的可信主体支持进行回顾和分析,提出了DLS(离散标记序列)多级安全模型.该模型将可信主体的生命周期分解为一系列非可信状态,对每一个状态赋予一个敏感标记.可信主体的当前敏感标记等于当前非可信状态的敏感标记,非可信状态的切换由预定义的可信请求事件触发.从而可信主体的当前敏感标记可以根据其应用逻辑而动态调整.同时给出了模型保持系统安全性的安全状态和规则.与Bell模型等可信主体敏感标记范围模型相比,该模型在多级安全的策略范围内实现了可信主体的特权最小化.

Analytical method of predicating the instabilities of a micro arch-shaped beam under electrostatic loading

An arch-shaped beam with different configurations under electrostatic loading experiences either the direct pull-in instability or the snap-through first and then the pull-in instability. When the pull-in instability occurs, the system collides with the electrode and adheres to it, which usually causes the system failure. When the snap-through instability occurs, the system experiences a discontinuous displacement to flip over without colliding with the electrode. The snap-through instability is an ideal actuation mechanism because of the following reasons: (1) after snap-through the system regains the stability and capability of withstanding further loading; (2) the system flips back when the loading is reduced, i.e. the system can be used repetitively; and (3) when approaching snap-through instability the system effective stiffness reduces toward zero, which leads to a fast flipping-over response. To differentiate these two types of instability responses for an arch-shaped beam is vital for the actuator design. For an arch-shaped beam under electrostatic loading, the nonlinear terms of the mid-plane stretching and the electrostatic loading make the analytical solution extremely difficult if not impossible and the related numerical solution is rather complex. Using the one mode expansion approximation and the truncation of the higher-order terms of the Taylor series, we present an analytical solution here. However, the one mode approximation and the truncation error of the Taylor series can cause serious error in the solution. Therefore, an error-compensating mechanism is also proposed. The analytical results are compared with both the experimental data and the numerical multi-mode analysis. The analytical method presented here offers a simple yet efficient solution approach by retaining good accuracy to analyze the instability of an arch-shaped beam under electrostatic loading.

Comparison of wavelet transform methods - The compression of IR spectral data

In this paper, an introduction of wavelet transform and multi-resolution analysis is presented. We describe three data compression methods based on wavelet transform for spectral information,and by using the multi-resolution analysis, we compressed spectral data by Daubechies's compactly supported orthogonal wavelet and orthogonal cubic B-splines wavelet, Using orthogonal cubic B-splines wavelet and coefficients of sharpening signal are set to zero, only very few large coefficients are stored, and a favourable data compression can be achieved.

节能有效的无线传感器网络状态信息收集算法

网络状态信息收集协议既要保证信息收集的准确性、实时性,又要保证协议算法的轻量级特性。为解决上述矛盾问题,提出了一种轻量级的、能量有效的、基于无损聚合的层次分簇数据收集机制(QTBDC)。QTBDC首先对网络节点编码并在节点间建立起一个逻辑层次簇结构,然后利用各个子簇状态数据的相似性和编码的连续性,实现了网内无损聚合。该监测机制使得网络状态信息的收集在不丢失数据细节信息的情况下,数据通信量大大减少。经过仿真分析表明,该方法与现有经典数据收集方法相比,实现了节能,延长了网络的生命期。

实现多数据源集成与互操作的扩展视图方法

文中针对多数据源集成与互操作提出了一种扩展视图方法．其基本思想是：以扩展对象模型（ＥＯＭ）为理论基础，通过扩展视图方法实现对多种数据源中的数据（对象）进行多层抽象与封装，以此满足分布环境下不同层次的应用集成需求．文中还讨论了扩展对象模型（ＥＯＭ）、基于扩展视图方法的集成机制，并结合具体的实例给出了扩展视图方法的形式化描述。

渤海海域河流相油藏精细描述技术及应用研究

Aiming at the character of Bohaii Sea area and the heterogeneity of fluvial facies reservoir, litho-geophysics experiments and integrated research of geophysical technologies are carried out. To deal with practical problems in oil fields of Bohai area, such as QHD32-6, Southern BZ25-1 and NP35-2 et al., technology of reservoir description based on seismic data and reservoir geophysical methods is built. In this dissertation, three points are emphasized: ①the integration of multidiscipline; ②the application of new methods and technologies; ③the integration of quiescent and dynamic data. At last, research of geology modeling and reservoir numerical simulation based on geophysical data are integrated. There are several innovative results and conclusion in this dissertation: (1)To deal with problems in shallow sea area where seismic data is the key data, a set of technologies for fine reservoir description based on seismic data in Bohai Sea area are built. All these technologies, including technologies of stratigraphic classification, sedimentary facies identification, structure fine characterization, reservoir description, fluid recognition and integration of geological modeling& reservoir numerical simulation, play an important role in the hydrocarbon exploration and development. In the research of lithology and hydrocarbon-bearing condition, petrophysical experiment is carried out. Outdoors inspection and experiment test data are integrated in seismic forward modeling& inversion research. Through the research, the seismic reflection rules of fluid in porosity are generated. Based on all the above research, seismic data is used to classify rock association, identify sedimentary facies belts and recognition hydrocarbon-bearing condition of reservoir. In this research, the geological meaning of geophysical information is more clear and the ambiguity of geophysical information is efficiently reduced, so the reliability in hydrocarbon forecasting is improved. The methods of multi-scales are developed in microfacies research aiming at the condition of shallow sea area in Bohai Sea: ① make the transformation from seismic information to sedimentary facies reality by discriminant analysis; ②in research of planar sedimentary facies, make microfacies research on seismic scale by technologies integration of seismic multi-attributes analysis& optimization, strata slicing and seismic waveform classification; ③descript the sedimentary facies distribution on scales below seismic resolution with the method of stochastic modeling. In the research of geological modeling and reservoir numerical simulation, the way of bilateral iteration between modeling and numerical simulation is carried out in the geological model correction. This process include several steps: ①make seismic forward modeling based on the reservoir numerical simulation results and geological models; ②get trend residual of forward modeling and real seismic data; ③make dynamic correction of the model according to the above trend residual. The modern integration technology of reservoir fine description research in Bohai Sea area, which is developed in this dissertation, is successfully used in (1)the reserve volume evaluation and development research in BZ25-1 oil field and (2)the tracing while drilling research in QHD32-6 oil field. These application researches show wide application potential in hydrocarbon exploration and development research in other oil fields.

基于独立分量分析的地震属性分析方法研究

Conventional seismic attribute analysis is not only time consuming, but also has several possible results. Therefore, seismic attribute optimization and multi-attribute analysis are needed. In this paper, Fuyu oil layer in Daqing oil field is our main studying object. And there is much difference between seismic attributes and well logs. So under this condition, Independent Component Analysis (ICA) and Kohonen neural net are introduced to seismic attribute optimization and multi-attribute analysis. The main contents are as follows: (1) Now the method of seismic attribute compression is mainly principal component analysis (PCA). In this article, independent component analysis (ICA), which is superficially related to PCA, but much more powerful, is used to seismic reservoir characterizeation. The fundamental, algorithms and applications of ICA are surveyed. And comparation of ICA with PCA is stydied. On basis of the ne-entropy measurement of independence, the FastICA algorithm is implemented. (2) Two parts of ICA application are included in this article: First, ICA is used directly to identify sedimentary characters. Combined with geology and well data, ICA results can be used to predict sedimentary characters. Second, ICA treats many attributes as multi-dimension random vectors. Through ICA transform, a few good new attributes can be got from a lot of seismic attributes. Attributes got from ICA optimization are independent. (3) In this paper, Kohonen self-organizing neural network is studied. First, the characteristics of neural network’s structure and algorithm is analyzed in detail, and the traditional algorithm is achieved which has been used in seism. From experimental results, we know that the Kohonen self-organizing neural network converges fast and classifies accurately. Second, the self-organizing feature map algorithm needs to be improved because the result of classification is not very exact, the boundary is not quite clear and the velocity is not fast enough, and so on. Here frequency sensitive principle is introduced. Combine it with the self-organizing feature map algorithm, then get frequency sensitive self-organizing feature map algorithm. Experimental results show that it is really better. (4) Kohonen self-organizing neural network is used to classify seismic attributes. And it can be avoided drawing confusing conclusions because the algorithm’s characteristics integrate many kinds of seismic features. The result can be used in the division of sand group’s seismic faces, and so on. And when attributes are extracted from seismic data, some useful information is lost because of difference and deriveative. But multiattributes can make this lost information compensated in a certain degree.