In industrial applications, the degradation rate of equipment is often accompanied by stochasticity due to constant changes in operating conditions and loads, making the degradation process tend to have a long-tailed distribution. When equipment fails rapidly due to untimely detection or excessive operation, it tends to produce extreme failure points and extend the tail, leading to the extreme long-tailed distribution. This situation makes it difficult for traditional RUL prediction methods to achieve satisfactory prediction and generalization performance. To address this challenge, this article proposes a new adaptive dynamic causal meta graph-task network for remaining useful life prediction with extreme long-tailed distribution condition. The model uses a data-driven approach in a dynamic graph framework to learn feature information with graph structure and dynamically update the pairwise dependencies of the learned latent features based on task changes. Then, we use causal analysis technique for inferring the distribution of latent representations of causal and outcome variables in the graph structure, which helps to infer the feature structure between latent variables. In addition, multiple balance factors independent of each other are introduced into the dynamic graph structure to adaptively capture meta-task and task-specific shared information, enhancing the model’s ability to capture the global task. Finally, in order to evaluate the proposed model, extensive experiments are conducted on engine and bearing degradation data, and the results validate the superiority of the model.

Identifying causal relationships from observational time-series data is a key problem in dealing with complex dynamical systems such as in the industrial or natural climate fields. Data-driven causal network construction in such systems is challenging since data sets are often high-dimensional and nonlinear. In response to this challenge, this paper combines partial rank correlation coefficients and proposes a new structure learning algorithm, TS-PRCS, suitable for time-series causal network models. In this article, we mainly make three contributions. First, we proved that partial rank correlation can be used as a standard of independence tests. Second, we combined partial rank correlation with constraint-based causality discovery methods, and proposed a causal network discovery algorithm (TS-PRCS) on time-series data based on partial rank correlation. Finally, the effectiveness of the algorithm is proven in experiments on time-series data generated by a time-series causal network model. Compared with an existing algorithm, the proposed algorithm achieves better results on high-dimensional and nonlinear data systems, and it also demonstrates good time performance. In particular, the algorithm has been applied to real data generated by a power plant. Experiments show that our method improves the ability to detect causality on time-series data, and further promotes the development of the field of causal network construction on time-series data.

Researchers recently started developing deep learning models capable of handling non-Euclidean data. However, because of existing framework limitations on model representations and learning algorithms, few have explored causal discovery on non-Euclidean data. This paper is the first attempt to do so. We start by proposing the Non-Euclidean Causal Model (NECM) which describes the causal generative relationship of non-Euclidean data and creates a new tensor data type along with a mapping process for the non-Euclidean causal mechanism. Second, within the NECM, we propose the non-Euclidean Hybrid Learning (NEHL) method, a causal discovery algorithm relying on the concept of the ball covariance recently introduced in the statistics field. Third, we generate two types of non-Euclidean datasets: Functional Data and Symmetric Positive Definite manifold data in conformity with the NECM. Finally, experimental results on the generated data and real-world data demonstrate the effectiveness of the proposed NEHL method.

Feature selection has been playing an important role in analyzing the high-dimension and low-sample-size gene expression profiles towards high classification performance of diseases and deep understanding of the underlying biological mechanisms. Besides classification performance, the stability of selected features is another non-ignorable factor in evaluating a feature selector, since stable feature selection results enhance the confidence of selected features for true biomarker discovery and further biological validation. In this study, we propose a novel feature selection method under the ensemble learning framework. Specifically, we take Fast Correlation Based Filter as the base feature selector to analyze subsamples of microarray data. We then present several aggregation methods to combine multiple feature subsets. Finally, two stability measures are used to quantify the robustness of feature selectors to data variations. Our comparative empirical study on publicly available datasets demonstrates the superiority of the proposed methods over its competitors in obtaining high stability scores and classification accuracy.

Computer-aided diagnosis (CAD) is mainly used in disease diagnosis and cause analysis. For example, using CAD to make early predictions of the semantic features of lung nodules is critical for helping physicians judge the semantic features of solitary pulmonary nodules. It is an effective method to predict disease using the features calculated from CT images. But how to select the most relevant features from the large number of image features is still a challenge. In this paper, we perform feature selection using maximum-relevance minimum-redundancy criteria based on applying a support vector machine (MRMR_SVM) on four types of computed image features to predict the semantic characteristics of pulmonary nodules over seven categories. The proposed method has the following advantages. 1) It improves work efficiency and reduces costs compared to manual evaluation by radiologists. 2) It combines a few key image features with specific semantic features to provide a basis for radiologists' diagnosis. 3) It eliminates noisy data and improves accuracy of early prediction of pulmonary nodules compared to using all features. The experimental results show that the proposed method performs well at predicting the semantic features of lung nodules in terms of accuracy and running time.

Microarray experiments on gene expression inevitably generate missing values, which impedes further downstream biological analysis. Therefore, it is key to estimate the missing values accurately. Most of the existing imputation methods tend to suffer from the over-fitting problem. In this study, we propose two regularized local learning methods for microarray missing value imputation. Motivated by the grouping effect of $L_{2}$L2 regularization, after selecting the target gene, we train an $L_{2}$L2 Regularized Local Least Squares imputation model (RLLSimpute_L2) on the target gene and its neighbors to estimate the missing values of the target gene. Furthermore, RLLSimpute_L2 imputes the missing values in an ascending order based on the associated missing rate with each target gene. This contributes to fully utilizing the previously estimated values. Besides $L_{2}$L2, we further explore $L_{1}$L1 regularization and propose an $L_{1}$L1 Regularized Local Least Squares imputation model (RLLSimpute_L1). To evaluate their effectiveness, we conducted extensive experimental studies on six benchmark datasets covering both time series and non-time series cases. Nine state-of-the-art imputation methods are compared with RLLSimpute_L2 and RLLSimpute_L1 in terms of three performance metrics. The comparative experimental results indicate that RLLSimpute_L2 outperforms its competitors by achieving smaller imputation errors and better structure preservation of differentially expressed genes.

A new algorithm, the Partial Correlation Statistic (PCS) algorithm, is presented for structure learning under linear Structural Equation Models. The PCS algorithm can deal with continuous data following linear arbitrary distribution rather than only a Gaussian distribution. This paper makes two specific contributions. First, for linear arbitrarily distributed datasets, which are generated by the linear structural equation models, if the sample size is sufficiently large, partial correlation coefficient statistic is proved to follow a Student's t-distribution. Second, the PCS algorithm combines hypothesis testing of partial correlation statistic and local learning to select potential neighbors of the target node. This significantly reduces the search space and achieves good time performance. The PCS algorithm does not need to choose optimal threshold of partial correlation by large amount of experiments. Especially, the PCS algorithm redefines the relevance from statistic theory and measure the relevance of the variables based on$p$-value. The effectiveness of the algorithm is compared with current state of the art methods on seven networks. A simulation shows that the PCS algorithm outperforms existing algorithms in terms of both accuracy and time performance on average.

Feature selection is a significant aspect of speech emotion recognition system. How to select a small subset out of the thousands of speech data is important for accurate classification of speech emotion. In this paper we investigate heuristic algorithm Harmony search (HS) for feature selection. We extract 3 feature sets, including MFCC, Fourier Parameters (FP), and features extracted with The Munich open Speech and Music Interpretation by Large Space Extraction (openSMILE) toolkit, from Berlin German emotion database (EMODB) and Chinese Elderly emotion database (EESDB). And combine MFCC with FP as the fourth feature set. We use Harmony search to select subsets and decrease the dimension space, and employ 10-fold cross validation in LIBSVM to evaluate the change of accuracy between selected subsets and original sets. Experimental results show that each subset's size reduced by about 50%, however, there is no sharp degeneration on accuracy and the accuracy almost maintains the original ones.

Gene selection plays a crucial role in the analysis of microarray data with high dimensionality and small sample size. Incremental wrapper based feature subset selection (FSS) methods, among various feature selection approaches, tend to obtain high quality feature subset and better classification accuracy than filter methods, while it is much more time consuming since the interdependence and redundancy between features is evaluated in a wrapper way. In this paper, we explore to introduce Markov Blanket (MB) into incremental wrapper based FSS process. Rather than evaluate the quality of all the features ranked by a filter method, our proposal eliminates features that are redundant to the newly selected one via MB during the wrapper evaluation process to reduce the number of wrappers, enabling us to select the relevant features and eliminate redundant ones efficiently. To verify the effectiveness and efficiency of the proposed approach, experimental comparisons on six publicly available microarray data are conducted with two typical classifiers with different metrics, Naïve Bayes and 1-Nearest-Neighbor. Experimental results demonstrate that our approach greatly speeds up the feature selection process, obtains more compact feature subset and achieves better classification accuracy compared to that without MB for both two-category and multi-category problems.

Correctly discovering causal relations from healthcare information can help people to understand disease mechanisms and discover disease causes. In some cases, the healthcare data do not follow a multivariate Gaussian distribution. We design a new causal structure learning algorithm. The algorithm can effectively combines ideas from local learning with simultaneous equations models techniques. In the first phase of the algorithm we select potential neighbors for each variable based on simultaneous equations models, and then perform a constrained hill-climbing search to orient the edges. Using the algorithm without prior knowledge, we analyze causal relations in the real data from the National Health and Nutrition Examination Survey.