Robust finite mixture regression for heterogeneous targets

Finite Mixture Regression (FMR) refers to the mixture modeling scheme which learns multiple regression models from the training data set. Each of them is in charge of a subset. FMR is an effective scheme for handling sample heterogeneity, where a single regression model is not enough for capturing the complexities of the conditional distribution of the observed samples given the features. In this paper, we propose an FMR model that (1) finds sample clusters and jointly models multiple incomplete mixed-type targets simultaneously, (2) achieves shared feature selection among tasks and cluster components, and (3) detects anomaly tasks or clustered structure among tasks, and accommodates outlier samples. We provide non-asymptotic oracle performance bounds for our model under a high-dimensional learning framework. The proposed model is evaluated on both synthetic and real-world data sets. The results show that our model can achieve state-of-the-art performance.     

基于空时变换网络的视频摘要生成

视频摘要生成是计算机视觉领域必不可少的关键任务,这一任务的目标是通过选择视频内容中信息最丰富的部分来生成一段简洁又完整的视频摘要,从而对视频内容进行总结.所生成的视频摘要通常为一组有代表性的视频帧（如视频关键帧）或按时间顺序将关键视频片段缝合所形成的一个较短的视频.虽然视频摘要生成方法的研究已经取得了相当大的进展,但现有的方法存在缺乏时序信息和特征表示不完备的问题,很容易影响视频摘要的正确性和完整性.为了解决视频摘要生成问题,本文提出一种空时变换网络模型,该模型包括三大模块,分别为：嵌入层、特征变换与融合层、输出层.其中,嵌入层可同时嵌入空间特征和时序特征,特征变换与融合层可实现多模态特征的变换和融合,最后输出层通过分段预测和关键镜头选择完成视频摘要的生成.通过空间特征和时序特征的分别嵌入,以弥补现有模型对时序信息表示的不足;通过多模态特征的变换和融合,以解决特征表示不完备的问题.我们在两个基准数据集上做了充分的实验和分析,验证了我们模型的有效性.     

Sparse probit linear mixed model

Linear mixed models (LMMs) are important tools in statistical genetics. When used for feature selection, they allow to find a sparse set of genetic traits that best predict a continuous phenotype of interest, while simultaneously correcting for various confounding factors such as age, ethnicity and population structure. Formulated as models for linear regression, LMMs have been restricted to continuous phenotypes. We introduce the sparse probit linear mixed model (Probit-LMM), where we generalize the LMM modeling paradigm to binary phenotypes. As a technical challenge, the model no longer possesses a closed-form likelihood function. In this paper, we present a scalable approximate inference algorithm that lets us fit the model to high-dimensional data sets. We show on three real-world examples from different domains that in the setup of binary labels, our algorithm leads to better prediction accuracies and also selects features which show less correlation with the confounding factors.     

A new approach for EEG signal classification of schizophrenic and control participants

This paper is concerned with a two stage procedure for analysis and classification of electroencephalogram (EEG) signals for twenty schizophrenic patients and twenty age-matched control participants. For each case, 20 channels of EEG are recorded. First, the more informative channels are selected using the mutual information techniques. Then, genetic programming is employed to select the best features from the selected channels. Several features including autoregressive model parameters, band power and fractal dimension are used for the purpose of classification. Both linear discriminant analysis (LDA) and adaptive boosting (Adaboost) are trained using tenfold cross validation to classify the reduced feature set and a classification accuracy of 85.90% and 91.94% is obtained by LDA and Adaboost, respectively. Another interesting observation from the channel selection procedure is that most of the selected channels are located in the prefrontal and temporal lobes confirming neuropsychological and neuroanatomical findings. The results obtained by the proposed approach are compared with a one stage procedure, the principal component analysis (PCA)-based feature selection, utilizing only 100 features selected from all channels. It is illustrated that the two stage procedure consisting of channel selection followed by feature reduction gives a more enhanced results in an efficient computation time.     

Using middle level features for robust shape tracking

Shape tracking with low level features (e.g., edge points) often fails in complex environments (e.g., in the presence of clutter, inner edges, or multiple objects). Two alternative methods are discussed in this paper. Both methods use middle level features (data centroids, strokes) which are more informative and reliable than edge transitions used in most tracking algorithms. Furthermore, it is assumed in this paper that each feature can be either a valid measurement or an outlier. A confidence degree is assigned to each feature or to a given interpretation of all visual features. Features/interpretations with high degrees of confidence have large influence on the shape estimates while features/interpretations with low degrees of confidence have negligible influence on the shape estimates. It is shown in this paper that both items (middle level features and confidence degrees) lead to a significant improvement of the tracker robustness and performance in the presence of clutter and abrupt shape and motion changes.     

Latent representation learning based autoencoder for unsupervised feature selection in hyperspectral imagery

In hyperspectral image (HSI) analysis, high-dimensional data may contain noisy, irrelevant and redundant information. To mitigate the negative effect from these information, feature selection is one of the useful solutions. Unsupervised feature selection is a data preprocessing technique for dimensionality reduction, which selects a subset of informative features without using any label information. Different from the linear models, the autoencoder is formulated to nonlinearly select informative features. The adjacency matrix of HSI can be constructed to extract the underlying relationship between each data point, where the latent representation of original data can be obtained via matrix factorization. Besides, a new feature representation can be also learnt from the autoencoder. For a same data matrix, different feature representations should consistently share the potential information. Motivated by these, in this paper, we propose a latent representation learning based autoencoder feature selection (LRLAFS) model, where the latent representation learning is used to steer feature selection for the autoencoder. To solve the proposed model, we advance an alternative optimization algorithm. Experimental results on three HSI datasets confirm the effectiveness of the proposed model.

     

Interpolating optimizing process control

A new model-based optimizing controller for a set of nonlinear systems is proposed. The nonlinear model set is based on a convex combination of two bounding linear models. An optimal control sequence is computed for each of the two bounding models. The proposed control algorithm is based on a convex combination of the two control sequences. A novel feature in these two optimizations is an added constraint related to the feasibility of the ‘other’ bounding model. The control algorithm can for example be used in model predictive control. We provide robust feasibility guarantees and an upper bound on the optimal criterion if the bounding models are linear FIR models. Further, simulation examples demonstrate significant feasibility improvements in the case where the bounding models are general linear state-space models. The proposed method guarantees robust feasibility for a 1-step ahead prediction in the general case. This can be of interest in MPC applications.     

结合特征图谱学习的人数统计方法

针对实际公共场景视频的人数统计中存在的背景干扰、光照变化、目标间遮挡等问题,提出一种结合特征图谱学习和一阶动态线性回归的人数统计方法。首先,建立图像的尺度不变特征变换（SIFT）特征与目标真实密度图之间的特征图谱映射模型,利用SIFT特征和前述映射模型得到包含目标和背景特征量的特征图谱;然后,根据通常监控视频中背景变化较小、特征图谱中的背景特征量相对稳定的特点,由特征图谱的积分与真实人数通过一阶动态线性回归建立人数回归模型;最后,通过该回归模型模型得出估计人数。在数据集MALL和PETS2009上进行实验,实验结果表明：与累积属性空间方法相比,所提方法平均绝对误差降低了2.2%;与基于角点检测的一阶动态线性回归方法相比,其平均绝对误差降低了6.5%,平均相对误差降低了2.3%。     

基于径向基函数的多目标回归特征构建算法

多目标回归（MTR）是一种针对单个样本同时具有多个连续型输出的回归问题。现有的多目标回归算法都基于同一个特征空间学习回归模型,而忽略了各输出目标本身的特殊性质。针对这一问题,提出基于径向基函数的多目标回归特征构建算法。首先,将各目标的输出作为额外的特征对各输出目标进行聚类,根据聚类中心在原始特征空间构成了目标特定特征空间的基;然后,通过径向基函数将原始特征空间映射到目标特定特征空间,构造目标特定的特征,并基于这些目标特定特征构建各输出目标的基回归模型;最后,用基回归模型的输出组成隐藏空间,采用低秩学习算法在其中发掘和利用输出目标之间的关联。在18个多目标回归数据集上进行实验,并把所提算法与层叠单目标回归（SST）、回归器链集成（ERC）、多层、多目标回归（MMR）等经典的多目标回归算法进行对比,结果表明所提算法在14个数据集上都取得了最好的性能,并且在18个数据集上的平均性能排序居第一位。可见所提算法构建的目标特定特征能够提高各输出目标的预测准确性,并结合低秩学习得到输出目标间的关联性以从整体上提升多目标回归的预测性能。     

Term-Dependent Confidence Normalisation for Out-of-Vocabulary Spoken Term Detection

An important component of a spoken term detection (STD) system involves estimating confidence measures of hypothesised detections.A potential problem of the widely used lattice-based confidence estimation,however,is that the confidence scores are treated uniformly for all search terms,regardless of how much they may differ in terms of phonetic or linguistic properties.This problem is particularly evident for out-of-vocabulary (OOV) terms which tend to exhibit high intra-term diversity.To address the impact of term diversity on confidence measures,we propose in this work a term-dependent normalisation technique which compensates for term diversity in confidence estimation.We first derive an evaluation-metric-oriented normalisation that optimises the evaluation metric by compensating for the diverse occurrence rates among terms,and then propose a linear bias compensation and a discriminative compensation to deal with the bias problem that is inherent in lattice-based confidence measurement and from which the Term Specific Threshold (TST) approach suffers.We tested the proposed technique on speech data from the multi-party meeting domain with two state-ofthe-art STD systems based on phonemes and words respectively.The experimental results demonstrate that the confidence normalisation approach leads to a significant performance improvement in STD,particularly for OOV terms with phonemebased systems.     

Multi-Target Tracking by Online Learning a CRF Model of Appearance and Motion Patterns

We introduce an online learning approach for multi-target tracking. Detection responses are gradually associated into tracklets in multiple levels to produce final tracks. Unlike most previous approaches which only focus on producing discriminative motion and appearance models for all targets, we further consider discriminative features for distinguishing difficult pairs of targets. The tracking problem is formulated using an online learned CRF model, and is transformed into an energy minimization problem. The energy functions include a set of unary functions that are based on motion and appearance models for discriminating all targets, as well as a set of pairwise functions that are based on models for differentiating corresponding pairs of tracklets. The online CRF approach is more powerful at distinguishing spatially close targets with similar appearances, as well as in tracking targets in presence of camera motions. An efficient algorithm is introduced for finding an association with low energy cost. We present results on four public data sets, and show significant improvements compared with several state-of-art methods.     

Statistical nonlinear analysis for reliable promotion decision-making

New economic conditions have led to innovations in retail industries, such as more dynamic retail approaches based on flexible strategies. We propose and compare different approaches incorporating nonlinear methods for promotional decision-making using retail aggregated data registered at the point of the sale. Specifically, this paper describes a reliable quantification tool as an effective information system leveraged on recent and historical data that provides managers with an operative vision. Furthermore, a new set of indicators are proposed to evaluate the reliability and stability of the data model in the multidimensional feature space by using nonparametric resampling techniques. This allows the user to make a clearer comparison among linear, nonlinear, static, and dynamic data models, and to identify the uncertainty of different feature space regions, for example, corresponding to the most frequent deal features. This methodology allows retailers to use aggregated data in suitable conditions that will result in acceptable confidence intervals. To test the proposed methodology, we used a database containing the sales history of representative products registered by a Spanish retail chain. The results indicate that: (1) the deal effect curve analysis and the time series linear model do not provide enough expressive capacity, and (2) nonlinear promotional models more accurately follow the actual sales pattern obtained in response to the implemented sales promotions. The quarterly temporal analysis conducted enabled the authors to identify long-term changes in the dynamics of the model for several products, especially during the early stage of most recent economic crisis, consistent with the information provided by the reliability indices in terms of the feature space. We conclude that the proposed method provides a reliable operative tool for decision support, allowing retailers to alter their strategies to accommodate consumer behavior.     

On transforming statistical models for non-frontal face verification

We address the pose mismatch problem which can occur in face verification systems that have only a single (frontal) face image available for training. In the framework of a Bayesian classifier based on mixtures of gaussians, the problem is tackled through extending each frontal face model with artificially synthesized models for non-frontal views. The synthesis methods are based on several implementations of maximum likelihood linear regression (MLLR), as well as standard multi-variate linear regression (LinReg). All synthesis techniques rely on prior information and learn how face models for the frontal view are related to face models for non-frontal views. The synthesis and extension approach is evaluated by applying it to two face verification systems: a holistic system (based on PCA-derived features) and a local feature system (based on DCT-derived features). Experiments on the FERET database suggest that for the holistic system, the LinReg-based technique is more suited than the MLLR-based techniques; for the local feature system, the results show that synthesis via a new MLLR implementation obtains better performance than synthesis based on traditional MLLR. The results further suggest that extending frontal models considerably reduces errors. It is also shown that the local feature system is less affected by view changes than the holistic system; this can be attributed to the parts based representation of the face, and, due to the classifier based on mixtures of gaussians, the lack of constraints on spatial relations between the face parts, allowing for deformations and movements of face areas.     

Feature-Based Image Analysis

According to Marr's paradigm of computational vision the first process is an extraction of relevant features. The goal of this paper is to quantify and characterize the information carried by features using image-structure measured at feature-points to reconstruct images. In this way, we indirectly evaluate the concept of feature-based image analysis. The main conclusions are that (i) a reasonably low number of features characterize the image to such a high degree, that visually appealing reconstructions are possible, (ii) different feature-types complement each other and all carry important information. The strategy is to define metamery classes of images and examine the information content of a canonical least informative representative of this class. Algorithms for identifying these are given. Finally, feature detectors localizing the most informative points relative to different complexity measures derived from models of natural image statistics, are given.     

利用多核增强学习的立体图像舒适度评价模型

目的 传统的立体视觉舒适度评价模型,在学习阶段一般采用回归算法,且需要大量的包含主观测试数据的训练样本,针对这个问题,提出一种利用多核增强学习分类算法的立体图像舒适度评价模型。方法 首先,考虑人们在实际观测图像时,对于先后观测到的不同图像进行相互比较的情况,将评价模型看成是偏好分类器,构造包含偏好标签的偏好立体图像对（PSIP）,构成PSIP训练集;其次,提取多个视差统计特征和神经学模型响应特征;然后,利用基于AdaBoost的多核学习算法来建立偏好标签与特征之间的关系模型,并分析偏好分类概率（即相对舒适度概率）与最终的视觉舒适度之间的映射关系。结果 在独立立体图像库上,与现有代表性回归算法相比较,本文算法的Pearson线性相关系数（PLCC）在0.84以上,Spearman等级相关系数（SRCC）在0.80以上,均优于其他模型的各评价指标;而在跨库测试中,本文算法的PLCC、SRCC指标均优于传统的支持向量回归算法。结论 相比于传统的回归算法,本文算法具有更好的评价性能,能够更为准确地预测立体图像视觉舒适度。     

The construction of fuzzy least squares estimators in fuzzy linear regression models

A new concept and method of imposing imprecise (fuzzy) input and output data upon the conventional linear regression model is proposed. Under the considerations of fuzzy parameters and fuzzy arithmetic operations (fuzzy addition and multiplication), we propose a fuzzy linear regression model which has the similar form as that of conventional one. We conduct the h-level (conventional) linear regression models of fuzzy linear regression model for the sake of invoking the statistical techniques in (conventional) linear regression analysis for real-valued data. In order to determine the sign (nonnegativity or nonpositivity) of fuzzy parameters, we perform the statistical testing hypotheses and evaluate the confidence intervals. Using the least squares estimators obtained from the h-level linear regression models, we can construct the membership functions of fuzzy least squares estimators via the form of “Resolution Identity” which is well-known in fuzzy sets theory. In order to obtain the membership degree of any given estimate taken from the fuzzy least squares estimator, optimization problems have to be solved. We also provide two computational procedures to deal with those optimization problems.     

SimiVal,a multi-criteria map comparison tool for land-change model projections

The multiple uses of land-cover models have led to validation with choice metrics or an ad hoc choice of the validation metrics available. To address this, we have identified the major dimensions of land-cover maps that ought to be evaluated and devised a Similarity Validation (SimiVal) tool. SimiVal uses a linear regression to test a modelled projection against benchmark cases of, perfect, observed and systematic-bias, calculated by rescaling the metrics from a random case relative to the observed, perfect case. The most informative regression coefficients, p-value and slope, are plot on a ternary graph of ‘similarity space’ whose extremes are the three benchmark cases. SimiVal is tested on projections of two deliberately contrasting land-cover models to show the similarity between intra- and inter-model parameterisations. We find metrics of landscape structure are important in distinguishing between different projections of the same model. Predictive and exploratory models can benefit from the tool.     

A hybrid of deep learning and hand-crafted features based approach for snow cover mapping

Monitoring the extent of snow cover plays a vital role for a better understanding of current and future climatic, ecological, and water cycle conditions. Previously, several traditional machine learning models have been applied for accomplishing this while exploring a variety of feature extraction techniques on various information sources. However, the laborious process of any amount of hand-crafted feature extraction has not helped to obtain high accuracies. Recently, deep learning models have shown that feature extraction can be made automatic and that they can achieve the required high accuracies but at the cost of requiring a large amount of labelled data. Fortunately, despite the absence of such large amounts of labelled data for this task, we can rely on pre-trained models, which accept red-green-blue (RGB) information (or dimensions-reduced spectral data). However, it is always better to include a variety of information sources to solve any problem, especially with the availability of other important information sources like synthetic aperture radar (SAR) imagery and elevation. We propose a hybrid model where the deep learning is assisted by these information sources which have until now been left out. Particularly, our model learns from both the deep learning features (derived from spectral data) and the hand-crafted features (derived from SAR and elevation). Such an approach shows interesting performance-improvement from 96.02% (through deep learning alone) to 98.10% when experiments were conducted for Khiroi village of the Himalayan region in India.     

Translating Braille into Chinese based on improved CBHG model

Transforming Mandarin Braille to Chinese text is a significant but less focused machine translation task. CBHG is a building block used in the Tacotron text-to-speech model. Since Mandarin Braille is constructed from the pronunciation of Chinese characters, CBHG can be used to perform Braille–Chinese translation. Unfortunately, only relying on the convolution blocks in CBHG cannot effectively extract the features of Braille sequences. Two ways are proposed to improve the CBHG model: CBHG-SE and CBHG-ECA. The two modules adaptively recalibrate channel-wise feature responses by explicitly modeling interdependencies between channels in CBHG. The quality of representations produced by the network can also be improved. Meanwhile, the network can learn to use global information to emphasize informative features and suppress less useful ones selectively. CBHG-ECA has stronger feature recalibration capabilities than CBHG-SE due to its more direct correspondence between channels and their weights. These two models can achieve 92.23 BLEU and 91.48 BLEU on the Braille–Chinese dataset, outperforming CBHG and other neural machine translation models.