支持向量机及其在径流预测中的应用

给出了支持向量机方法(SVM)的思路、特点及关键之处,探讨了SVM在径流预测中的可能性,并与基于遗传算法的门限回归模型(TR) 进行了对比分析。径流预测实例分析表明,在拟合阶段,SVM模型要好于TR模型;在预留检验阶段,SVM模型与TR模型接近。同时SVM模型适合于小样本情况且能达到全局最优。SVM模型用于径流预测是可行的、优越的。     

基于支持向量机的旋转机组状态趋势预示技术

本文主要介绍了支持向量机在旋转机组状态趋势预示中的运用。通过对某旋转机组的振动烈度进行预测,并将其结果与使用时间序列进行预测的结果相比较,发现使用支持向量机进行预测的结果更好。     

TESTING CHANGE-POINTS IN THE EXPLOSIVE GAUSSIAN AUTOREGRESSIVE PROCESSES

Abstract. A functional limit theorem with a particular function class and topology is derived for non-ergodic type time series. This limit theorem allows us to study the asymptotic law of the associated likelihood ratio test (LRT) statistic for testing the presence of a change in the covariance parameter in the explosive Gaussian autoregressive model. We show that the level of the LRT cannot be approximated without introducing appropriate normalization. The limit law of a particular weighted likelihood ratio test is examined through a simulation study and is compared with the well-known Kolmogorov distribution obtained in the stationary case; we conclude that for practical applications when the root is really close to unity one can use the same thresholds as in the stationary case. This procedure is applied to the study of three real time series known to be non-stationary.     

基于新型混合模型的欠定盲分离方法

针对欠定盲分离问题,提出了一种新的源恢复方法. 在时频域局部区域采用复高斯分布对源信号进行建模,将语音信号的稀疏性和局部平稳性结合在一起,提出了一种新的混合模型来描述观测信号在局部区域的概率分布.通过该模型,将每个时频点的源信号状态的判断问题转换成模型的参数估计和后验概率的计算问题,最后通过子混合矩阵的逆恢复出源信号. 实验结果表明,该方法具有很快的收敛速度,并且比已有方法具有更好的分离性能.     

基于运动相关皮层电位握力运动模式识别研究

面向基于脑-机接口（Brain-computer interface,BCI）的脑-机交互控制（Brain-machine interaction control,BMIC）——直接脑控机器人,提出一种新的左、右手握力运动参数范式,在该范式下探索左、右手握力运动相关皮层电位/运动相关电位（Movement-related potentials,MRPs）的时域特征表示并识别握力运动模式.在涉及左、右手4个不同任务的实验中采集了11个健康被试的脑电信号,任务期间要求被试以2种握力变化模式之一完成自愿握力运动,每种任务随机重复30次.不同握力任务之间具有显著差异的运动相关电位特征用于识别握力运动模式.分别用基于核的Fisher线性判别分析和支持向量机识别4个不同的握力运动任务.研究结果进一步证实运动相关电位可以表征握力运动规划、运动执行和运动监控的脑神经机制过程.基于核的Fisher线性判别分析和支持向量机分别获得24±4%和21±5%的平均错误分类率.最小误分类率是12%,所有被试平均最小误分类率为20.9±5%.与传统的仅仅识别参与运动的肢体类型以及识别单侧肢体运动参数的研究相比,本研究可望为脑-机交互控制/脑控机器人接口提供更多的力控制意图指令,奠定了后续的对比研究基础.     

Data-mining based SQL injection attack detection using internal query trees

Detecting SQL injection attacks (SQLIAs) is becoming increasingly important in database-driven web sites. Until now, most of the studies on SQLIA detection have focused on the structured query language (SQL) structure at the application level. Unfortunately, this approach inevitably fails to detect those attacks that use already stored procedure and data within the database system. In this paper, we propose a framework to detect SQLIAs at database level by using SVM classification and various kernel functions. The key issue of SQLIA detection framework is how to represent the internal query tree collected from database log suitable for SVM classification algorithm in order to acquire good performance in detecting SQLIAs. To solve the issue, we first propose a novel method to convert the query tree into an n-dimensional feature vector by using a multi-dimensional sequence as an intermediate representation. The reason that it is difficult to directly convert the query tree into an n-dimensional feature vector is the complexity and variability of the query tree structure. Second, we propose a method to extract the syntactic features, as well as the semantic features when generating feature vector. Third, we propose a method to transform string feature values into numeric feature values, combining multiple statistical models. The combined model maps one string value to one numeric value by containing the multiple characteristic of each string value. In order to demonstrate the feasibility of our proposals in practical environments, we implement the SQLIA detection system based on PostgreSQL, a popular open source database system, and we perform experiments. The experimental results using the internal query trees of PostgreSQL validate that our proposal is effective in detecting SQLIAs, with at least 99.6% of the probability that the probability for malicious queries to be correctly predicted as SQLIA is greater than the probability for normal queries to be incorrectly predicted as SQLIA. Finally, we perform additional experiments to compare our proposal with syntax-focused feature extraction and single statistical model based on feature transformation. The experimental results show that our proposal significantly increases the probability of correctly detecting SQLIAs for various SQL statements, when compared to the previous methods.     

FREQUENCY DOMAIN TESTS OF MULTIVARIATE GAUSSIANITY AND LINEARITY

A stationary multivariate time series { X _t } is defined as linear if it can be written in the form X _t = ∑^∞ _{j =−∞} A _j e _{t − j} where A _j are square matrices and e _t are independent and identically distributed random vectors. If the e _t } are normally distributed, then { X _t is a multivariate Gaussian linear process. This paper is concerned with the testing of departures of a vector stationary process from multivariate Gaussianity and linearity using the bispectral approach. First the definition and properties of cumulants of random matrices are used to obtain the expressions for the higher-order cumulant and spectral vectors of a linear vector process as defined above. Then it is shown that linearity of a vector process implies constancy of the modulus square of its normalized higher-order spectra whereas the component of such a vector process does not necessarily have a linear representation. Finally, statistics for the testing of multivariate Gaussianity and linearity are proposed.     

Exploring album structure for face recognition in online social networks

In this paper, we propose an album-oriented face-recognition model that exploits the album structure for face recognition in online social networks. Albums, usually associated with pictures of a small group of people at a certain event or occasion, provide vital information that can be used to effectively reduce the possible list of candidate labels. We show how this intuition can be formalized into a model that expresses a prior on how albums tend to have many pictures of a small number of people. We also show how it can be extended to include other information available in a social network. Using two real-world datasets independently drawn from Facebook, we show that this model is broadly applicable and can significantly improve recognition rates.     

A study of process monitoring based on inverse Gaussian distribution

The inverse Gaussian distribution has considerable applications in describing product life, employee service times, and so on. In this paper, the average run length (ARL) unbiased control charts, which monitor the shape and location parameters of the inverse Gaussian distribution respectively, are proposed when the in-control parameters are known. The effects of parameter estimation on the performance of the proposed control charts are also studied. An ARL-unbiased control chart for the shape parameter with the desired ${\mathit{ARL}}_0$, which takes the variability of the parameter estimate into account, is further developed. The performance of the proposed control charts is investigated in terms of the ARL and standard deviation of the run length. Finally, an example is used to illustrate the proposed control charts.     

Efficient data selection for speech recognition based on prior confidence estimation using speech and monophone models

This paper proposes an efficient speech data selection technique that can identify those data that will be well recognized. Conventional confidence measure techniques can also identify well-recognized speech data. However, those techniques require a lot of computation time for speech recognition processing to estimate confidence scores. Speech data with low confidence should not go through the time-consuming recognition process since they will yield erroneous spoken documents that will eventually be rejected. The proposed technique can select the speech data that will be acceptable for speech recognition applications. It rapidly selects speech data with high prior confidence based on acoustic likelihood values and using only speech and monophone models. Experiments show that the proposed confidence estimation technique is over 50 times faster than the conventional posterior confidence measure while providing equivalent data selection performance for speech recognition and spoken document retrieval.