Continuous speech recognition based on ICA and geometrical learning

We investigate the use of independent component analysis (ICA) for speech feature extraction in digits speech recognition systems. We observe that this may be true for recognition tasks based on Geometrical Learning with little training data. In contrast to image processing, phase information is not essential for digits speech recognition. We therefore propose a new scheme that shows how the phase sensitivity can be removed by using an analytical description of the ICA-adapted basis functions. Furthermore, since the basis functions are not shift invariant, we extend the method to include a frequency-based ICA stage that removes redundant time shift information. The digits speech recognition results show promising accuracy. Experiments show that the method based on ICA and Geometrical Learning outperforms HMM in a different number of training samples.

Application of ICA in on-line verification of the phase difference of the current sensor

The performance of the current sensor in power equipment may become worse affected by the environment. In this paper, based on ICA, we propose a method for on-line verification of the phase difference of the current sensor. However, not all source components are mutually independent in our application. In order to get an exact result, we have proposed a relative likelihood index to choose an optimal result from different runs. The index is based on the maximum likelihood evaluation theory and the independent subspace analysis. The feasibility of our method has been confirmed by experimental results.

Continuous speech recognition based on ICA and geometrical learning

We investigate the use of independent component analysis (ICA) for speech feature extraction in digits speech recognition systems. We observe that this may be true for recognition tasks based on Geometrical Learning with little training data. In contrast to image processing, phase information is not essential for digits speech recognition. We therefore propose a new scheme that shows how the phase sensitivity can be removed by using an analytical description of the ICA-adapted basis functions. Furthermore, since the basis functions are not shift invariant, we extend the method to include a frequency-based ICA stage that removes redundant time shift information. The digits speech recognition results show promising accuracy. Experiments show that the method based on ICA and Geometrical Learning outperforms HMM in a different number of training samples.

一种基于新型数据预测方法的MICA仿真研究

在多元统计过程监控中,为解决因未知过程数据统计分布而产生误报漏报的现象,提出一种结合多向独立元分析法(MICA)和广义相关系数(GCC)数据预测的综合方法,进行在线监控过程的仿真。MICA分析方法能有效分解各变量的关联关系,且不需考虑建模数据是否符合正态分布,用此方法计算的独立元变量能更好地描述过程的变化规律。为提高预报未来过程故障的能力,提出用广义相关系数法进行数据预测:确定与运行轨迹相似的监控模型库中的轨迹,并使其相应部分承接于运行轨迹之后。现场采集聚氯乙烯聚合过程的数据进行仿真,仿真结果显示:对于在线监控和在线故障诊断方面,这种新型预测方法优于其它传统处理预测问题的方法。

基于独立成分分析和模糊支持向量机的人脸识别方法

成分分析(PCA)只能从2阶上去消除数据的相关性，传统支持向量机在解决多类问题时会出现分类的盲区问题，针对这两种情况，首先采用独立成分分析(ICA)方法解决了高阶上的数据相关性问题；同时在传统支持向量机中引入模糊隶属度函数，用模糊支持向量机解决了传统支持向量机在多类数据识别中的盲区问题。通过实验证明了该方法在人脸识别率上取得了显著提高。

基于改进SICA的人脸特征提取方法研究

独立分量分析是一种有效的人脸特征提取方法。考虑到人脸样本的对称性,本文采用对称独立分量分析的方法对人脸样本进行特征提取。为了提高独立分量分析法表征人脸特征空间的能力,本文采用遗传算法对特征空间进行选择优化,获得最优的人脸特征子集。仿真实验表明,本文提出方法的识别率明显的好于独立分量分析方法的识别率。

基于可复制性的独立成分分析——应用于功能磁共振成像数据

This dissertation systematically depicted and improved the application of Independent Component Analysis (ICA) to Functional Magnetic Resonance Imaging (fMRI), following the logic of verification, improvement, extension, and application. The concept of “reproducibility” was the philosophy throughout its four concluded studies. In the “verification” study, ICA was applied to the resting-state fMRI data, verified the resultant components with reproducibility, and examined the consistency of the results from ICA and traditional “seed voxel” method. At the meantime, the limitation of ICA application on fMRI data analysis was presented. In the “improvement” study, an improved ICA algorithm based on reproducibility, RAICAR, was developed to aid some of the limitations of ICA application. RAICAR was able to rank ICA components by reproducibility, determine the number of reliable components, and obtain more stable results. RAICAR provided useful tools for validation and interpretation of ICA results. In the “extension” study, RAICAR as well as the concept of “reproducibility” was extended to multi-subject ICA analysis, and gRAICAR algorithm was developed. gRAICAR allows some variation across subjects, examining common components among subjects. gRAICAR is also capable to detect potential subject grouping on some components. It is a new way for exploratory group analysis on fMRI. In the “application” study, two newly developed methods, RAICAR and gRAICAR, were used to investigate the effect of early music training on the brain mechanism of memory and learning. The results showed brain mechanism difference in memory retrieval and learning process between two groups of subjects. This study also verified the usefulness and importance of the new methods.

An analysis on overall crack-number-density of short-fatigue-cracks

The evolution of dispersed short-fatigue-cracks is analysed based on the equilibrium of crack-number-density (CND). By separating the mean value and the stochastic fluctuation of local CND, the equilibrium equation of overall CND is derived. Comparing with the mean-field equilibrium equation, the equilibrium equation of overall CND has different forms in the expression of crack-nucleation-rate or crack-growth-rate. The simulation results are compared with experimental measurements showing the stochastic analyses provide consistent tendency with experiments. The discrepancy in simulation results between overall CND and mean-field CND is discussed.

Higher-order analysis of crack-tip fields in power-law hardening materials

In this paper, we present an exact higher-order asymptotic analysis on the near-crack-tip fields in elastic-plastic materials under plane strain, Mode I. A four- or five-term asymptotic series of the solutions is derived. It is found that when 1.6 < n less-than-or-equal-to 2.8 (here, n is the hardening exponent), the elastic effect enters the third-order stress field; but when 2.8< n less-than-or-equal-to 3.7 this effect turns to enter the fourth-order field, with the fifth-order field independent. Moreover, if n>3.7, the elasticity only affects the fields whose order is higher than 4. In this case, the fourth-order field remains independent. Our investigation also shows that as long as n is larger than 1.6, the third-order field is always not independent, whose amplitude coefficient K3 depends either on K1 or on both K1 and K2 (K1 and K2 arc the amplitude coefficients of the first- and second-order fields, respectively). Firmly, good agreement is found between our results and O'Dowd and Shih's numerical ones[8] by comparison.

Analysis on the spin reorientation of GdFeCo/DyFeCo exchange-coupled double-layer films

In this paper, GdFeCo/DyFeCo exchange-coupled double-layer films used for center aperture type magnetically induced super resolution were investigated through experiments and theoretical calculation. The samples were prepared by magnetron sputtering method. The polar Kerr effect was measured to prove the spin reorientation of the readout layer. Theoretical study of magnetization profiles was performed on the basis of the mean-field theory and the continuum model. The theoretical results showed that the magnetization orientation of the readout layer changed gradually from in-plane to out-of-plane with the rise of the temperature. Theoretical analysis explained the experimental results successfully. (c) 2005 Elsevier B.V. All rights reserved.

Electromagnetic-field-induced cohesive force enhancement in a metallic nanowire

We theoretically study the conducting electronic contribution to the cohesive force in a metallic nanowire irradiated under a transversely polarized external electromagnetic field at low temperatures and in the ballistic regime. In the framework of the free-electron model, we have obtained a time-dependent two-level electronic wavefunction by means of a unitary transformation. Using a thermodynamic statistical approach with this wavefunction, we have calculated the cohesive force in the nanowire. We show that the cohesive force can be divided into two components, one of which is independent of the electromagnetic field (static component), which is consistent with the existing results in the literature. The magnitude of the other component is proportional to the electromagnetic field strength. This extra component of the cohesive force is originally from the coherent coupling between the two lateral energy levels of the wire and the electromagnetic field.

Equivalent-barotropic definition of tropospheric mean temperature

An equivalent-barotropic (EB) description of the tropospheric temperature field is derived from the geostrophic empirical mode (GEM) in the form of a scalar function Gamma(p, phi), where p is pressure and phi is 300-850-mb thickness. Baroclinic parameter phi plays the role of latitude at each longitudinal section. Compared with traditional Eulerian-mean methods, GEM defines a mean field in baroclinic streamfunction space with a time scale much longer than synoptic variability. It prompts an EB concept that is only based on a baroclinic field. Monthly GEM fields are diagnosed from NCEP-NCAR reanalysis data and account for more than 90% of the tropospheric thermal variance. The circumglobal composite of GEM fields exhibits seasonal, zonal, and hemispheric asymmetries, with larger rms errors occurring in winter and in the Northern Hemisphere (NH). Zonally asymmetric features and planetary deviation from EB are seen in the NH winter GEM. Reconstruction of synoptic sections and correlation analysis reveal that the tropospheric temperature field is EB at the leading order and has a 1-day phase lag behind barotropic variations in extratropical regions.

Damage Localization, Sensitivity of Energy Release and the Catastrophe Transition

Large earthquakes can be viewed as catastrophic ruptures in the earth’s crust. There are two common features prior to the catastrophe transition in heterogeneous media. One is damage localization and the other is critical sensitivity; both of which are related to a cascade of damage coalescence. In this paper, in an attempt to reveal the physics underlying the catastrophe transition, analytic analysis based on mean-field approximation of a heterogeneous medium as well as numerical simulations using a network model are presented. Both the emergence of damage localization and the sensitivity of energy release are examined to explore the inherent statistical precursors prior to the eventual catastrophic rupture. Emergence of damage localization, as predicted by the mean-field analysis, is consistent with observations of the evolution of damage patterns. It is confirmed that precursors can be extracted from the time-series of energy release according to its sensitivity to increasing crustal stress. As a major result, present research indicates that the catastrophe transition and the critical point hypothesis (CPH) of earthquakes are interrelated. The results suggest there may be two cross-checking precursors of large earthquakes: damage localization and critical sensitivity.

Relationship between strain localization and catastrophic rupture

In order to explore a prior warning to catastrophic rupture of heterogeneous media, like rocks, the present study investigates the relationship between surface strain localization and catastrophic rupture. Instrumented observations on the evolution of surface strain field and the catastrophic rupture of a rock under uniaxial compression were carried out. It is found that the evolution of surface strain field displays two phases: at the early stage, the strain field keeps nearly uniform with weak fluctuations increasing slowly; but at the stage prior to catastrophic rupture, a certain accelerating localization develops and a localized zone emerges. Based on the measurements, an analysis was performed with local mean-field approximation. More importantly, it is found that the scale of localized zone is closely related to the catastrophic rupture strain and the rupture strain can be calculated in accord with the local-mean-field model satisfactorily. This provides a possible clue to the forecast of catastrophic rupture. (c) 2007 Elsevier Ltd. All rights reserved.

Elastic and statistically brittle (ESB) model, damage localization and catastrophic rupture

In this paper, an elastic and statistically brittle (ESB) model is applied to the process of damage evolution induced catastrophic rupture and the influence of localization and softening on catastrophic rupture is discussed. According to the analysis, the uncertainty of catastrophic rupture should be attributed to the unknown scale of localized zone. Based on the elastic and statistically brittle model but local mean field approximation, the relation between the scale of localized zone and catastrophic rupture is obtained and then justified with experiments. These results can not only give a deeper understanding of the mechanism governing catastrophic rupture, but also provide a possible tool to foresee the occurrence of catastrophic rupture.