Mutual information-based optimization of sparse spatio-spectral filters in brain–computer interface

Recently, neuro-rehabilitation based on brain–computer interface (BCI) has been considered one of the important applications for BCI. A key challenge in this system is the accurate and reliable detection of motor imagery. In motor imagery-based BCIs, the common spatial patterns (CSP) algorithm is widely used to extract discriminative patterns from electroencephalography signals. However, the CSP algorithm is sensitive to noise and artifacts, and its performance depends on the operational frequency band. To address these issues, this paper proposes a novel optimized sparse spatio-spectral filtering (OSSSF) algorithm. The proposed OSSSF algorithm combines a filter bank framework with sparse CSP filters to automatically select subject-specific discriminative frequency bands as well as to robustify against noise and artifacts. The proposed algorithm directly selects the optimal regularization parameters using a novel mutual information-based approach, instead of the cross-validation approach that is computationally intractable in a filter bank framework. The performance of the proposed OSSSF algorithm is evaluated on a dataset from 11 stroke patients performing neuro-rehabilitation, as well as on the publicly available BCI competition III dataset IVa. The results show that the proposed OSSSF algorithm outperforms the existing algorithms based on CSP, stationary CSP, sparse CSP and filter bank CSP in terms of the classification accuracy, and substantially reduce the computational time of selecting the regularization parameters compared with the cross-validation approach.     

最优区域共空间模式的运动想象脑电信号分类方法

共空间模式(Common Spatial Pattern,CSP)是脑机接口(Brain-Computer Interface,BCI)中一种有效的特征提取方法,然而传统CSP算法并未考虑在提取前剔除可能会影响其性能的不相关的嘈杂通道信号。所以针对不同对象的通道选择问题,提出了一种最优区域共空间模式(ORCSP)特征提取方法。首先通过欧式距离得到每个通道的附近区域,再根据方差比选择可分性最高的区域,然后采用5折交叉验证对区域内通道数目进行寻优,进而得到区分度最高的区域特征,最后使用支持向量机(SVM)进行分类。所提方法在BCI竞赛数据上进行了实验测试,并与同类型的正则化CSP和局部区域CSP算法进行了对比,在BCI Competition Ⅲ Dataset Ⅳ a数据集上达到了89.78%的平均准确率。实验结果验证了所提出方法的有效性。     

Class mean embedding for face recognition

Recently, local discriminant embedding (LDE) was proposed as a means of addressing manifold learning and pattern classification. In the LDE framework, the neighbor and class of data points are used to construct the graph embedding for classification problems. From a high dimensional to a low dimensional subspace, data points of the same class maintain their intrinsic neighbor relations, whereas neighboring data points of different classes no longer stick to one another. But, neighboring data points of different classes are not deemphasized efficiently by LDE and it may degrade the performance of classification. In this paper, we investigate its extension, called class mean embedding (CME), using class mean of data points to enhance its discriminant power in their mapping into a low dimensional space. After joined class mean data points, (1) CME may cause each class of data points to be more compact in the high dimension space; (2) CME may increase the quantity of data points, and solves the small sample size (SSS) problem; (3) CME may preserve well the local geometry of the data manifolds in the embedding space. Experimental results on ORL, Yale, AR, and FERET face databases show the effectiveness of the proposed method.     

Graph embedding and extensions: a general framework for dimensionality reduction

A large family of algorithms - supervised or unsupervised; stemming from statistics or geometry theory - has been designed to provide different solutions to the problem of dimensionality reduction. Despite the different motivations of these algorithms, we present in this paper a general formulation known as graph embedding to unify them within a common framework. In graph embedding, each algorithm can be considered as the direct graph embedding or its linear/kernel/tensor extension of a specific intrinsic graph that describes certain desired statistical or geometric properties of a data set, with constraints from scale normalization or a penalty graph that characterizes a statistical or geometric property that should be avoided. Furthermore, the graph embedding framework can be used as a general platform for developing new dimensionality reduction algorithms. By utilizing this framework as a tool, we propose a new supervised dimensionality reduction algorithm called marginal Fisher analysis in which the intrinsic graph characterizes the intraclass compactness and connects each data point with its neighboring points of the same class, while the penalty graph connects the marginal points and characterizes the interclass separability. We show that MFA effectively overcomes the limitations of the traditional linear discriminant analysis algorithm due to data distribution assumptions and available projection directions. Real face recognition experiments show the superiority of our proposed MFA in comparison to LDA, also for corresponding kernel and tensor extensions     

基于判别正则化的局部保留投影

局部保留投影（Locality preserving projections,LPP）是一种常用的线性化流形学习方法,其通过线性嵌入来保留基于图所描述的流形数据本质结构特征,因此LPP对图的依赖性强,且在嵌入过程中缺少对图描述的进一步分析和挖掘。当图对数据本质结构特征描述不恰当时,LPP在嵌入过程中不易实现流形数据本质结构的有效提取。为了解决这个问题,本文在给定流形数据图描述的条件下,通过引入局部相似度阈值进行局部判别分析,并据此建立判别正则化局部保留投影（简称DRLPP）。该方法能够在现有图描述的条件下,有效突出不同流形结构在线性嵌入空间中的可分性。在人造合成数据集和实际标准数据集上对DRLPP以及相关算法进行对比实验,实验结果证明了DRLPP的有效性。     

Adaptive Graph Embedding Discriminant Projections

Graph embedding based learning method plays an increasingly significant role on dimensionality reduction (DR). However, the selection to neighbor parameters of graph is intractable. In this paper, we present a novel DR method called adaptive graph embedding discriminant projections (AGEDP). Compared with most existing DR methods based on graph embedding, such as marginal Fisher analysis which usually predefines the intraclass and interclass neighbor parameters, AGEDP applies all the homogeneous samples for constructing the intrinsic graph, and simultaneously selects heterogeneous samples within the neighborhood generated by the farthest homogeneous sample for constructing the penalty graph. Therefore, AGEDP not only greatly enhances the intraclass compactness and interclass separability, but also adaptively performs neighbor parameter selection which considers the fact that local manifold structure of each sample is generally different. Experiments on AR and COIL-20 datasets demonstrate the effectiveness of the proposed method for face recognition and object categorization, and especially under the interference of occlusion, noise and poses, it is superior to other graph embedding based methods with three different classifiers: nearest neighbor classifier, sparse representation classifier and linear regression classifier.     

Neural network based auto association and time-series prediction for biosignal processing in brain-computer interfaces

Neural networks (NNs) can be deployed in many different ways in signal processing applications. This paper illustrates how neural networks are employed in a prediction based preprocessing framework, referred to as neural-time-series-prediction-preprocessing (NTSPP), in an electroencephalogram (EEG)-based brain-computer interface (BCI). NTSPP has been shown to increase feature separability by mapping the original EEG signals via time-series-prediction to a higher dimensional space. Preliminary results of a similar novel framework are also presented where, instead of using predictive NNs, auto-associative NNs are employed and features are extracted from the output of auto-associative NNs trained to specialize on EEG signals for particular brain states. The results show that this preprocessing framework referred to as auto-associative NN preprocessing (ANNP) also has the potential to improve the performance of BCIs. Both the NTSPP and ANNP are compared with and deployed in conjunction with the well know common spatial patterns (CSP) to produce a BCI system which significantly outperforms either approach operating independently and has the potential to produce good performances even with a lower number of EEG channels compared to a multichannel BCI. Multichannel BCIs normally perform better that 2-3 channel BCIs however reducing the number of EEG channels required can positively impact on the time needed to mount electrodes and minimize the obtrusiveness of the electrode montage for the user. It is also shown that NTSPP can improve the potential for employing existing BCI methods with minimal subject-specific parameter tuning to deploy the BCI autonomously. Results are presented with six different classification approaches including various statistical classifiers such as Linear Discriminant Analysis (LDA), Support Vector Machines (SVM) and a Bayes classifier.     

提升图嵌入框架及在表情识别中的应用 *

提出了一种提升图嵌入框架用于特征提取和选择 ,以及一种新的近邻权重计算方法 ,称为分类图。传统图嵌入模型的近邻权重采用欧氏距离 ,不能被提升算法所更新 ;相比较 ,分类图采用的是提升算法中样本的权重,反映的是样本在分类过程中的重要程度 ,有效地提高了图嵌入模型的分类性能。在通用人脸表情库上的识别实验结果验证了提升图嵌入模型的有效性。     

运动想象脑电的块选择共空间模式特征提取

共空间模式(CSP)作为一种空间滤波方法已在脑电信号(EEG)的特征提取上得到了广泛应用,而对脑电信号的通道和频带进行合理选择可以有效改善共空间模式特征在运动想象脑机接口(BCI)中的分类性能.针对已有选择方法中未充分考虑通道间差异性的问题,本文提出一种对通道和频带同时进行选择的块选择共空间模式(BS–CSP)特征提取...     

基于Beta分布和半监督学习的非确定性知识图谱嵌入模型

近年来,面向确定性知识图谱的嵌入模型在知识图谱补全等任务中取得了长足的进展,但如何设计和训练面向非确定性知识图谱的嵌入模型仍然是一个重要挑战。不同于确定性知识图谱,非确定性知识图谱的每个事实三元组都有着对应的置信度,因此,非确定性知识图谱嵌入模型需要准确地计算出每个三元组的置信度。现有的非确定性知识图谱嵌入模型结构较为简单,只能处理对称关系,并且无法很好地处理假负(false-negative)样本问题。为了解决上述问题,该文首先提出了一个用于训练非确定性知识图谱嵌入模型的统一框架,该框架使用基于多模型的半监督学习方法训练非确定性知识图谱嵌入模型。为了解决半监督学习中半监督样本噪声过高的问题,我们还使用蒙特卡洛Dropout计算出模型对输出结果的不确定度,并根据该不确定度有效地过滤了半监督样本中的噪声数据。此外,为了更好地表示非确定性知识图谱中实体和关系的不确定性以处理更复杂的关系,该文还提出了基于Beta分布的非确定性知识图谱嵌入模型UBetaE,该模型将实体、关系均表示为一组相互独立的Beta分布。在公开数据集上的实验结果表明,结合该文所提出的半监督学习方法和UBetaE模型,不仅...     

多类运动想象脑电信号的两级特征提取方法

共同空间模式（Common spatial pattern,CSP）是运动想象脑机接口（Brain-computer interface,BCI）中常用的特征提取方法,但对多类任务的分类正确率却明显低于两类任务.通过引入堆叠降噪自动编码器（Stacked denoising autoencoders,SDA）,提出了一种多类运动想象脑电信号（Electroencephalogram,EEG）的两级特征提取方法.首先利用一对多CSP（One versus rest CSP,OVR-CSP）将脑电信号变换到使信号方差区别最大的低维空间,然后通过SDA网络提取其中可以更好表达类别属性的高层抽象特征,最后使用Softmax分类器进行分类.在对BCI竞赛IV中Data-sets 2a的4类运动想象任务进行的分类实验中,平均Kappa系数达到0.69,表明了所提出的特征提取方法的有效性和鲁棒性.     

Semi-supervised dimension reduction based on hypergraph embedding for hyperspectral images

Dimension reduction (DR) is an efficient and effective preprocessing step of hyperspectral images (HSIs) classification. Graph embedding is a frequently used model for DR, which preserves some geometric or statistical properties of original data set. The embedding using simple graph only considers the relationship between two data points, while in real-world application, the complex relationship between several data points is more important. To overcome this problem, we present a linear semi-supervised DR method based on hypergraph embedding (SHGE) which is an improvement of semi-supervised graph learning (SEGL). The proposed SHGE method aims to find a projection matrix through building a semi-supervised hypergraph which can preserve the complex relationship of the data and the class discrimination for DR. Experimental results demonstrate that our method achieves better performance than some existing DR methods for HSIs classification and is time saving compared with the existed method SEGL which used simple graph.     

Feature extraction based on fuzzy class mean embedding (FCME) with its application to face and palm biometrics

In the local discriminant embedding (LDE) framework, the neighbor and class of data points were used to construct the graph embedding for classification problems. From a high-dimensional to a low-dimensional subspace, data points of the same class maintain their intrinsic neighbor relations, whereas neighboring data points of different classes no longer stick to one another. However, face images are always affected by variations in illumination conditions and different facial expressions in the real world. So, distant data points are not deemphasized efficiently by LDE and it may degrade the performance of classification. In order to solve above problems, in this paper, we investigate the fuzzy set theory and class mean of LDE, called fuzzy class mean embedding (FCME), using the fuzzy k-nearest neighbor (FKNN) and the class sample average to enhance its discriminant power in their mapping into a low dimensional space. In the proposed method, a membership degree matrix is firstly calculated using FKNN, then the membership degree and class mean are incorporated into the definition of the Laplacian scatter matrix. The optimal projections of FCME can be obtained by solving a generalized eigenfunction. Experimental results on the Wine dataset, ORL, Yale, AR, FERET face database and PolyU palmprint database show the effectiveness of the proposed method.     

基于自编码器的贝叶斯网嵌入及概率推理

贝叶斯网(BN)是不确定性知识表示和推理的基本框架,广泛用于社交网络、知识图谱和医疗诊断等领域.特定领域中基于BN的分析诊断和决策支持,其核心计算任务是基于BN进行多次概率推理.然而,使用传统的概率推理方法,基于同一BN的多次概率推理其中间过程存在很多重复的计算结果,具有较高的时间复杂度.为了提高多次概率推理的效率,提出易于重用和易于计算的贝叶斯网嵌入及相应的概率推理方法.首先,借鉴图嵌入的基本思想,使用点互信息矩阵来表示BN的有向无环图结构和条件概率参数,提出基于自编码器和注意力机制的BN嵌入方法.其中,自编码器的每一编码层利用节点与其邻居节点(父节点和子节点)的相关性生成节点嵌入,从而在嵌入向量中保存BN节点间的概率依赖关系.然后,使用嵌入向量之间的距离来度量节点之间的联合概率,提出基于嵌入向量的BN概率推理方法.实验证明,针对BN的多次概率推理,所提方法的效率高于现有方法,且能得到准确的推理结果.     

An extended EM algorithm for joint feature extraction and classification in brain-computer interfaces

For many electroencephalogram (EEG)-based brain-computer interfaces (BCIs), a tedious and time-consuming training process is needed to set parameters. In BCI Competition 2005, reducing the training process was explicitly proposed as a task. Furthermore, an effective BCI system needs to be adaptive to dynamic variations of brain signals; that is, its parameters need to be adjusted online. In this article, we introduce an extended expectation maximization (EM) algorithm, where the extraction and classification of common spatial pattern (CSP) features are performed jointly and iteratively. In each iteration, the training data set is updated using all or part of the test data and the labels predicted in the previous iteration. Based on the updated training data set, the CSP features are reextracted and classified using a standard EM algorithm. Since the training data set is updated frequently, the initial training data set can be small (semi-supervised case) or null (unsupervised case). During the above iterations, the parameters of the Bayes classifier and the CSP transformation matrix are also updated concurrently. In online situations, we can still run the training process to adjust the system parameters using unlabeled data while a subject is using the BCI system. The effectiveness of the algorithm depends on the robustness of CSP feature to noise and iteration convergence, which are discussed in this article. Our proposed approach has been applied to data set IVa of BCI Competition 2005. The data analysis results show that we can obtain satisfying prediction accuracy using our algorithm in the semisupervised and unsupervised cases. The convergence of the algorithm and robustness of CSP feature are also demonstrated in our data analysis.     

Feature Extraction Using a Complete Kernel Extension of Supervised Graph Embedding

In this article, the kernel-based methods explained by a graph embedding framework are analyzed and their nature is revealed, i.e. any kernel-based method in a graph embedding framework is equivalent to kernel principal component analysis plus its corresponding linear one. Based on this result, the authors propose a complete kernel-based algorithms framework. Any algorithm in our framework makes full use of two kinds of discriminant information, irregular and regular. The proposed algorithms framework is tested and evaluated using the ORL, Yale and FERET face databases. The experiment results demonstrate the effectiveness of our proposed algorithms framework.     

Making FLDA applicable to face recognition with one sample per person

In face recognition, the Fisherface approach based on Fisher linear discriminant analysis (FLDA) has obtained some success. However, FLDA fails when each person just has one training face sample available because of nonexistence of the intra-class scatter. In this paper, we propose to partition each face image into a set of sub-images with the same dimensionality, therefore obtaining multiple training samples for each class, and then apply FLDA to the set of newly produced samples. Experimental results on the FERET face database show that the proposed approach is feasible and better in recognition performance than E(PC)²A.     

DISCOVERING ROBUST EMBEDDINGS IN (DIS)SIMILARITY SPACE FOR HIGH‐DIMENSIONAL LINGUISTIC FEATURES

Recent research has shown the effectiveness of rich feature representation for tasks in natural language processing (NLP). However, exceedingly large number of features do not always improve classification performance. They may contain redundant information, lead to noisy feature presentations, and also render the learning algorithms intractable. In this paper, we propose a supervised embedding framework that modifies the relative positions between instances to increase the compatibility between the input features and the output labels and meanwhile preserves the local distribution of the original data in the embedded space. The proposed framework attempts to support flexible balance between the preservation of intrinsic geometry and the enhancement of class separability for both interclass and intraclass instances. It takes into account characteristics of linguistic features by using an inner product‐based optimization template. (Dis)similarity features, also known as empirical kernel mapping, is employed to enable computationally tractable processing of extremely high‐dimensional input, and also to handle nonlinearities in embedding generation when necessary. Evaluated on two NLP tasks with six data sets, the proposed framework provides better classification performance than the support vector machine without using any dimensionality reduction technique. It also generates embeddings with better class discriminability as compared to many existing embedding algorithms.     

Harmonic Mean of Kullback–Leibler Divergences for Optimizing Multi-Class EEG Spatio-Temporal Filters

The common spatial patterns (CSP) is a classical approach to spatial filtering of electroencephalogram (EEG) used in brain-computer interfaces. The local temporal common spatial patterns (LTCSP) method is a temporal generalization of CSP. Both CSP and LTCSP, however, are only suitable for the two-class paradigm. In this paper, we address this limitation under the framework of Kullback?CLeibler (KL) divergence. We show that CSP is equivalent to maximizing the symmetric KL divergence of two class-conditional probability density functions under the Gaussian assumption. This analysis establishes a probabilistic interpretation for CSP, as well as LTCSP. Based on the KL formulation, we propose a new multi-class extension to optimizing the spatio-temporal filters by maximizing the harmonic mean of all pairs of symmetric KL divergences between the filtered class-conditional densities. Experiments of classification of multiple EEG classes on the data sets of BCI competition show the effectiveness of the proposed methods.     

Graph embedding discriminant analysis for face recognition

This paper develops a supervised discriminant technique, called graph embedding discriminant analysis (GEDA), for dimensionality reduction of high-dimensional data in small sample size problems. GEDA can be seen as a linear approximation of a multimanifold-based learning framework in which nonlocal property is taken into account besides the marginal property and local property. GEDA seeks to find a set of perfect projections that not only can impact the samples of intraclass and maximize the margin of interclass, but also can maximize the nonlocal scatter at the same time. This characteristic makes GEDA more intuitive and more powerful than linear discriminant analysis (LDA) and marginal fisher analysis (MFA). The proposed method is applied to face recognition and is examined on the Yale, ORL and AR face image databases. The experimental results show that GEDA consistently outperforms LDA and MFA when the training sample size per class is small.