Search

In deep learning of classifiers, the cost function usually takes the form of a combination of SoftMax and CrossEntropy functions. The SoftMax unit transforms the scores predicted by the model network into assessments of the degree (probabilities) of an object's membership to a given class. On the other hand, CrossEntropy measures the divergence of this prediction from the distribution of target scores. This work introduces the ISBE functionality, justifying the thesis about the redundancy of cross-entropy computation in deep learning of classifiers. Not only can we omit the calculation of entropy, but also, during back-propagation, there is no need to direct the error to the normalization unit for its backward transformation. Instead, the error is sent directly to the model's network. Using examples of perceptron and convolutional networks as classifiers of images from the MNIST collection, it is observed for ISBE that results are not degraded with SoftMax only but also with other activation functions such as Sigmoid, Tanh, or their hard variants HardSigmoid and HardTanh. Moreover, savings in the total number of operations were observed within the forward and backward stages. The article is addressed to all deep learning enthusiasts but primarily to programmers and students interested in the design of deep models. For example, it illustrates in code snippets possible ways to implement ISBE functionality but also formally proves that the SoftMax trick only applies to the class of dilated SoftMax functions with relocations. [ABSTRACT FROM AUTHOR]

The properties of cardio-respiratory coupling (CRC) are affected by various pathological conditions related to the cardiovascular and/or respiratory systems. In heart failure, one of the most common cardiac pathological conditions, the degree of CRC changes primarily depend on the type of heart-rhythm alterations. In this work, we investigated CRC in heart-failure patients, applying measures from information theory, i.e., Granger Causality (GC), Transfer Entropy (TE) and Cross Entropy (CE), to quantify the directed coupling and causality between cardiac (RR interval) and respiratory (Resp) time series. Patients were divided into three groups depending on their heart rhythm (sinus rhythm and presence of low/high number of ventricular extrasystoles) and were studied also after cardiac resynchronization therapy (CRT), distinguishing responders and non-responders to the therapy. The information-theoretic analysis of bidirectional cardio-respiratory interactions in HF patients revealed the strong effect of nonlinear components in the RR (high number of ventricular extrasystoles) and in the Resp time series (respiratory sinus arrhythmia) as well as in their causal interactions. We showed that GC as a linear model measure is not sensitive to both nonlinear components and only model free measures as TE and CE may quantify them. CRT responders mainly exhibit unchanged asymmetry in the TE values, with statistically significant dominance of the information flow from Resp to RR over the opposite flow from RR to Resp, before and after CRT. In non-responders this asymmetry was statistically significant only after CRT. Our results indicate that the success of CRT is related to corresponding information transfer between the cardiac and respiratory signal quantified at baseline measurements, which could contribute to a better selection of patients for this type of therapy. [ABSTRACT FROM AUTHOR]

In deep learning of classifiers, the cost function usually takes the form of a combination of SoftMax and CrossEntropy functions. The SoftMax unit transforms the scores predicted by the model network into assessments of the degree (probabilities) of an object’s membership to a given class. On the other hand, CrossEntropy measures the divergence of this prediction from the distribution of target scores. This work introduces the ISBE functionality, justifying the thesis about the redundancy of cross-entropy computation in deep learning of classifiers. Not only can we omit the calculation of entropy, but also, during back-propagation, there is no need to direct the error to the normalization unit for its backward transformation. Instead, the error is sent directly to the model’s network. Using examples of perceptron and convolutional networks as classifiers of images from the MNIST collection, it is observed for ISBE that results are not degraded with SoftMax only but also with other activation functions such as Sigmoid, Tanh, or their hard variants HardSigmoid and HardTanh. Moreover, savings in the total number of operations were observed within the forward and backward stages. The article is addressed to all deep learning enthusiasts but primarily to programmers and students interested in the design of deep models. For example, it illustrates in code snippets possible ways to implement ISBE functionality but also formally proves that the SoftMax trick only applies to the class of dilated SoftMax functions with relocations.

Tri-training expands the training set by adding pseudo-labels to unlabeled data, which effectively improves the generalization ability of the classifier, but it is easy to mislabel unlabeled data into training noise, which damages the learning efficiency of the classifier, and the explicit decision mechanism tends to make the training noise degrade the accuracy of the classification model in the prediction stage. This study proposes the Tri-training algorithm for adaptive nearest neighbor density editing and cross-entropy evaluation (TTADEC), which is used to reduce the training noise formed during the classifier iteration and to solve the problem of inaccurate prediction by explicit decision mechanism. First, the TTADEC algorithm uses the nearest neighbor editing to label high-confidence samples. Then, combined with the relative nearest neighbor to define the local density of samples to screen the pre-training samples, and then dynamically expand the training set by adaptive technique. Finally, the decision process uses cross-entropy to evaluate the completed base classifier of training and assign appropriate weights to it to construct a decision function. The effectiveness of the TTADEC algorithm is verified on the UCI dataset, and the experimental results show that compared with the standard Tri-training algorithm and its improvement algorithm, the TTADEC algorithm has better classification performance and can effectively deal with the semi-supervised classification problem where the training set is insufficient. [ABSTRACT FROM AUTHOR]

In order to overcome the spectral interference of the conventional Fourier transform in the International Electrotechnical Commission framework, this paper introduces a Bregman-split-based compressive sensing (BSCS) method to estimate the Taylor–Fourier coefficients in a multi-frequency dynamic phasor model. Considering the DDC component estimation, this paper transforms the phasor problem into a compressive sensing model based on the regularity and sparsity of the dynamic harmonic signal distribution. It then derives an optimized hybrid regularization algorithm with the Bregman split method to reconstruct the dynamic phasor estimation. The accuracy of the model was verified by using the cross entropy to measure the distribution differences of values. Composite tests derived from the dynamic phasor test conditions were then used to verify the potentialities of the BSCS method. Simulation results show that the algorithm can alleviate the impact of dynamic signals on phasor estimation and significantly improve the estimation accuracy, which provides a theoretical basis for P-class phasor measurement units (PMUs). [ABSTRACT FROM AUTHOR]

Tri-training expands the training set by adding pseudo-labels to unlabeled data, which effectively improves the generalization ability of the classifier, but it is easy to mislabel unlabeled data into training noise, which damages the learning efficiency of the classifier, and the explicit decision mechanism tends to make the training noise degrade the accuracy of the classification model in the prediction stage. This study proposes the Tri-training algorithm for adaptive nearest neighbor density editing and cross-entropy evaluation (TTADEC), which is used to reduce the training noise formed during the classifier iteration and to solve the problem of inaccurate prediction by explicit decision mechanism. First, the TTADEC algorithm uses the nearest neighbor editing to label high-confidence samples. Then, combined with the relative nearest neighbor to define the local density of samples to screen the pre-training samples, and then dynamically expand the training set by adaptive technique. Finally, the decision process uses cross-entropy to evaluate the completed base classifier of training and assign appropriate weights to it to construct a decision function. The effectiveness of the TTADEC algorithm is verified on the UCI dataset, and the experimental results show that compared with the standard Tri-training algorithm and its improvement algorithm, the TTADEC algorithm has better classification performance and can effectively deal with the semi-supervised classification problem where the training set is insufficient.

In order to overcome the spectral interference of the conventional Fourier transform in the International Electrotechnical Commission framework, this paper introduces a Bregman-split-based compressive sensing (BSCS) method to estimate the Taylor–Fourier coefficients in a multi-frequency dynamic phasor model. Considering the DDC component estimation, this paper transforms the phasor problem into a compressive sensing model based on the regularity and sparsity of the dynamic harmonic signal distribution. It then derives an optimized hybrid regularization algorithm with the Bregman split method to reconstruct the dynamic phasor estimation. The accuracy of the model was verified by using the cross entropy to measure the distribution differences of values. Composite tests derived from the dynamic phasor test conditions were then used to verify the potentialities of the BSCS method. Simulation results show that the algorithm can alleviate the impact of dynamic signals on phasor estimation and significantly improve the estimation accuracy, which provides a theoretical basis for P-class phasor measurement units (PMUs).

In this paper, a joint early stopping criterion based on cross entropy (CE), named joint CE criterion, is presented for double-protograph low-density parity-check (DP-LDPC) codes-based joint source-channel coding (JSCC) systems in images transmission to reduce the decoding complexity and decrease the decoding delay. The proposed early stopping criterion adopts the CE from the output likelihood ratios (LLRs) of the joint decoder. Moreover, a special phenomenon named asymmetry oscillation-like convergence (AOLC) in the changing process of CE is uncovered in the source decoder and channel decoder of this system meanwhile, and the proposed joint CE criterion can reduce the impact from the AOLC phenomenon. Comparing to the counterparts, the results show that the joint CE criterion can perform well in the decoding complexity and decoding latency in the low–moderate signal-to-noise ratio (SNR) region and achieve performance improvement in the high SNR region with appropriate parameters, which also demonstrates that this system with joint CE is a low-latency and low-power system.

Heart sound signals reflect valuable information about heart condition. Previous studies have suggested that the information contained in single-channel heart sound signals can be used to detect coronary artery disease (CAD). But accuracy based on single-channel heart sound signal is not satisfactory. This paper proposed a method based on multi-domain feature fusion of multi-channel heart sound signals, in which entropy features and cross entropy features are also included. A total of 36 subjects enrolled in the data collection, including 21 CAD patients and 15 non-CAD subjects. For each subject, five-channel heart sound signals were recorded synchronously for 5 min. After data segmentation and quality evaluation, 553 samples were left in the CAD group and 438 samples in the non-CAD group. The time-domain, frequency-domain, entropy, and cross entropy features were extracted. After feature selection, the optimal feature set was fed into the support vector machine for classification. The results showed that from single-channel to multi-channel, the classification accuracy has increased from 78.75% to 86.70%. After adding entropy features and cross entropy features, the classification accuracy continued to increase to 90.92%. The study indicated that the method based on multi-domain feature fusion of multi-channel heart sound signals could provide more information for CAD detection, and entropy features and cross entropy features played an important role in it.

In recent years, on the basis of drawing lessons from traditional neural network models, people have been paying more and more attention to the design of neural network architectures for processing graph structure data, which are called graph neural networks (GNN). GCN, namely, graph convolution networks, are neural network models in GNN. GCN extends the convolution operation from traditional data (such as images) to graph data, and it is essentially a feature extractor, which aggregates the features of neighborhood nodes into those of target nodes. In the process of aggregating features, GCN uses the Laplacian matrix to assign different importance to the nodes in the neighborhood of the target nodes. Since graph-structured data are inherently non-Euclidean, we seek to use a non-Euclidean mathematical tool, namely, Riemannian geometry, to analyze graphs (networks). In this paper, we present a novel model for semi-supervised learning called the Ricci curvature-based graph convolutional neural network, i.e., RCGCN. The aggregation pattern of RCGCN is inspired by that of GCN. We regard the network as a discrete manifold, and then use Ricci curvature to assign different importance to the nodes within the neighborhood of the target nodes. Ricci curvature is related to the optimal transport distance, which can well reflect the geometric structure of the underlying space of the network. The node importance given by Ricci curvature can better reflect the relationships between the target node and the nodes in the neighborhood. The proposed model scales linearly with the number of edges in the network. Experiments demonstrated that RCGCN achieves a significant performance gain over baseline methods on benchmark datasets.

We address the problem of unsupervised anomaly detection for multivariate data. Traditional machine learning based anomaly detection algorithms rely on specific assumptions of normal patterns and fail to model complex feature interactions and relations. Recently, existing deep learning based methods are promising for extracting representations from complex features. These methods train an auxiliary task, e.g., reconstruction and prediction, on normal samples. They further assume that anomalies fail to perform well on the auxiliary task since they are never trained during the model optimization. However, the assumption does not always hold in practice. Deep models may also perform the auxiliary task well on anomalous samples, leading to the failure detection of anomalies. To effectively detect anomalies for multivariate data, this paper introduces a teacher-student distillation based framework Distillated Teacher-Student Network Ensemble (DTSNE). The paradigm of the teacher-student distillation is able to deal with high-dimensional complex features. In addition, an ensemble of student networks provides a better capability to avoid generalizing the auxiliary task performance on anomalous samples. To validate the effectiveness of our model, we conduct extensive experiments on real-world datasets. Experimental results show superior performance of DTSNE over competing methods. Analysis and discussion towards the behavior of our model are also provided in the experiment section.

QT interval variability (QTV) and heart rate variability (HRV) are both accepted biomarkers for cardiovascular events. QTV characterizes the variations in ventricular depolarization and repolarization. It is a predominant element of HRV. However, QTV is also believed to accept direct inputs from upstream control system. How QTV varies along with HRV is yet to be elucidated. We studied the dynamic relationship of QTV and HRV during different physiological conditions from resting, to cycling, and to recovering. We applied several entropy-based measures to examine their bivariate relationships, including cross sample entropy (XSampEn), cross fuzzy entropy (XFuzzyEn), cross conditional entropy (XCE), and joint distribution entropy (JDistEn). Results showed no statistically significant differences in XSampEn, XFuzzyEn, and XCE across different physiological states. Interestingly, JDistEn demonstrated significant decreases during cycling as compared with that during the resting state. Besides, JDistEn also showed a progressively recovering trend from cycling to the first 3 min during recovering, and further to the second 3 min during recovering. It appeared to be fully recovered to its level in the resting state during the second 3 min during the recovering phase. The results suggest that there is certain nonlinear temporal relationship between QTV and HRV, and that the JDistEn could help unravel this nuanced property.

Automatic identification of human interaction is a challenging task especially in dynamic environments with cluttered backgrounds from video sequences. Advancements in computer vision sensor technologies provide powerful effects in human interaction recognition (HIR) during routine daily life. In this paper, we propose a novel features extraction method which incorporates robust entropy optimization and an efficient Maximum Entropy Markov Model (MEMM) for HIR via multiple vision sensors. The main objectives of proposed methodology are: (1) to propose a hybrid of four novel features—i.e., spatio-temporal features, energy-based features, shape based angular and geometric features—and a motion-orthogonal histogram of oriented gradient (MO-HOG); (2) to encode hybrid feature descriptors using a codebook, a Gaussian mixture model (GMM) and fisher encoding; (3) to optimize the encoded feature using a cross entropy optimization function; (4) to apply a MEMM classification algorithm to examine empirical expectations and highest entropy, which measure pattern variances to achieve outperformed HIR accuracy results. Our system is tested over three well-known datasets: SBU Kinect interaction; UoL 3D social activity; UT-interaction datasets. Through wide experimentations, the proposed features extraction algorithm, along with cross entropy optimization, has achieved the average accuracy rate of 91.25% with SBU, 90.4% with UoL and 87.4% with UT-Interaction datasets. The proposed HIR system will be applicable to a wide variety of man–machine interfaces, such as public-place surveillance, future medical applications, virtual reality, fitness exercises and 3D interactive gaming.

This paper proposes a deep convolutional neural network model with encoder-decoder architecture to extract road network from satellite images. We employ ResNet-18 and Atrous Spatial Pyramid Pooling technique to trade off between the extraction precision and running time. A modified cross entropy loss function is proposed to train our deep model. A PointRend algorithm is used to recover a smooth, clear and sharp road boundary. The augmentated DeepGlobe dataset is used to train our deep model and the asynchronous training method is applied to accelerate the training process. Five salellite images covering Xiaomu village are taken as input to evaluate our model. The proposed E-Road model has fewer number of parameters and shorter training time. The experiments show E-Road outperforms other state-of-the-art deep models with 5.84% to 59.09% improvement, and can give the accurate predictions for the images with complex environment.

Embedding learning on knowledge graphs (KGs) aims to encode all entities and relationships into a continuous vector space, which provides an effective and flexible method to implement downstream knowledge-driven artificial intelligence (AI) and natural language processing (NLP) tasks. Since KG construction usually involves automatic mechanisms with less human supervision, it inevitably brings in plenty of noises to KGs. However, most conventional KG embedding approaches inappropriately assume that all facts in existing KGs are completely correct and ignore noise issues, which brings about potentially serious errors. To address this issue, in this paper we propose a novel approach to learn embeddings with triple trustiness on KGs, which takes possible noises into consideration. Specifically, we calculate the trustiness value of triples according to the rich and relatively reliable information from large amounts of entity type instances and entity descriptions in KGs. In addition, we present a cross-entropy based loss function for model optimization. In experiments, we evaluate our models on KG noise detection, KG completion and classification. Through extensive experiments on three datasets, we demonstrate that our proposed model can learn better embeddings than all baselines on noisy KGs.

The recognition of the voltage sag sources is the basis for formulating a voltage sag governance plan and clarifying the responsibility for the accident. Aiming at the recognition problem of voltage sag sources, a recognition method of voltage sag sources based on phase space reconstruction and improved Visual Geometry Group (VGG) transfer learning is proposed from the perspective of image classification. Firstly, phase space reconstruction technology is used to transform voltage sag signals, generate reconstruction images of voltage sag, and analyze the intuitive characteristics of different sag sources from reconstruction images. Secondly, combined with the attention mechanism, the standard VGG 16 model is improved to extract the features completely and prevent over-fitting. Finally, VGG transfer learning model uses the idea of transfer learning for training, which improves the efficiency of model training and the recognition accuracy of sag sources. The purpose of the training model is to minimize the cross entropy loss function. The simulation analysis verifies the effectiveness and superiority of the proposed method.

Rainfall is an essential index to measure drought, and it is dependent upon various parameters including geographical environment, air temperature and pressure. The nonlinear nature of climatic variables leads to problems such as poor accuracy and instability in traditional forecasting methods. In this paper, the combined forecasting method based on data mining technology and cross entropy is proposed to forecast the rainfall with full consideration of the time-effectiveness of historical data. In view of the flaws of the fuzzy clustering method which is easy to fall into local optimal solution and low speed of operation, the ant colony algorithm is adopted to overcome these shortcomings and, as a result, refine the model. The method for determining weights is also improved by using the cross entropy. Besides, the forecast is conducted by analyzing the weighted average rainfall based on Thiessen polygon in the Beijing-Tianjin-Hebei region. Since the predictive errors are calculated, the results show that improved ant colony fuzzy clustering can effectively select historical data and enhance the accuracy of prediction so that the damage caused by extreme weather events like droughts and floods can be greatly lessened and even kept at bay. [ABSTRACT FROM AUTHOR]

Cross entropy and Kullback⁻Leibler (K-L) divergence are fundamental quantities of information theory, and they are widely used in many fields. Since cross entropy is the negated logarithm of likelihood, minimizing cross entropy is equivalent to maximizing likelihood, and thus, cross entropy is applied for optimization in machine learning. K-L divergence also stands independently as a commonly used metric for measuring the difference between two distributions. In this paper, we introduce new inequalities regarding cross entropy and K-L divergence by using the fact that cross entropy is the negated logarithm of the weighted geometric mean. We first apply the well-known rearrangement inequality, followed by a recent theorem on weighted Kolmogorov means, and, finally, we introduce a new theorem that directly applies to inequalities between K-L divergences. To illustrate our results, we show numerical examples of distributions.

A consistent entropy estimator for hyperspherical data is proposed based on the k-nearest neighbor (knn) approach. The asymptotic unbiasedness and consistency of the estimator are proved. Moreover, cross entropy and Kullback-Leibler (KL) divergence estimators are also discussed. Simulation studies are conducted to assess the performance of the estimators for models including uniform and von Mises-Fisher distributions. The proposed knn entropy estimator is compared with the moment based counterpart via simulations. The results show that these two methods are comparable.

Accounting informatization is an important part of enterprise informatization. It affects the accounting and finance operational efficiency. With the comprehensive evaluation of accounting informatization from multiple aspects, we can find the strengths and weaknesses of corporate accounting informatization, which then can be improved precisely. In this paper, an evaluation approach of accounting informatization is proposed. Firstly, the evaluation index system is constructed from the aspects of strategic position, infrastructure construction, implementation of accounting informatization, informatization guarantee, and application efficiency. Considering the complexity and ambiguity of the index, experts are required to give linguistic ratings which are then converted into intuitionistic fuzzy number. Then, an entropy and cross entropy method based on intuitionistic fuzzy sets is proposed to derive the weights of experts so as to reduce the error caused by personal bias. By combining the weights of the index and weighted ratings, the evaluation results are obtained. Finally, a case of accounting information evaluation is given to illustrate the feasibility of the proposed approach.

In this study, we developed a model of combined streamflow forecasting based on cross entropy to solve the problems of streamflow complexity and random hydrological processes. First, we analyzed the streamflow data obtained from Wudaogou station on the Huifa River, which is the second tributary of the Songhua River, and found that the streamflow was characterized by fluctuations and periodicity, and it was closely related to rainfall. The proposed method involves selecting similar years based on the gray correlation degree. The forecasting results obtained by the time series model (autoregressive integrated moving average), improved grey forecasting model, and artificial neural network model (a radial basis function) were used as a single forecasting model, and from the viewpoint of the probability density, the method for determining weights was improved by using the cross entropy model. The numerical results showed that compared with the single forecasting model, the combined forecasting model improved the stability of the forecasting model, and the prediction accuracy was better than that of conventional combined forecasting models. [ABSTRACT FROM AUTHOR]

In this paper, we propose a novel and generic family of multiple importance sampling estimators. We first revisit the celebrated balance heuristic estimator, a widely used Monte Carlo technique for the approximation of intractable integrals. Then, we establish a generalized framework for the combination of samples simulated from multiple proposals. Our approach is based on considering as free parameters both the sampling rates and the combination coefficients, which are the same in the balance heuristics estimator. Thus our novel framework contains the balance heuristic as a particular case. We study the optimal choice of the free parameters in such a way that the variance of the resulting estimator is minimized. A theoretical variance study shows the optimal solution is always better than the balance heuristic estimator (except in degenerate cases where both are the same). We also give sufficient conditions on the parameter values for the new generalized estimator to be better than the balance heuristic estimator, and one necessary and sufficient condition related to χ 2 divergence. Using five numerical examples, we first show the gap in the efficiency of both new and classical balance heuristic estimators, for equal sampling and for several state of the art sampling rates. Then, for these five examples, we find the variances for some notable selection of parameters showing that, for the important case of equal count of samples, our new estimator with an optimal selection of parameters outperforms the classical balance heuristic. Finally, new heuristics are introduced that exploit the theoretical findings. [ABSTRACT FROM AUTHOR]

A consistent entropy estimator for hyperspherical data is proposed based on the κ-nearest neighbor (knn) approach. The asymptotic unbiasedness and consistency of the estimator are proved. Moreover, cross entropy and Kullback-Leibler (KL) divergence estimators are also discussed. Simulation studies are conducted to assess the performance of the estimators for models including uniform and von Mises-Fisher distributions. The proposed knn entropy estimator is compared with the moment based counterpart via simulations. The results show that these two methods are comparable. [ABSTRACT FROM AUTHOR]

In this paper, a joint early stopping criterion based on cross entropy (CE), named joint CE criterion, is presented for double-protograph low-density parity-check (DP-LDPC) codes-based joint source-channel coding (JSCC) systems in images transmission to reduce the decoding complexity and decrease the decoding delay. The proposed early stopping criterion adopts the CE from the output likelihood ratios (LLRs) of the joint decoder. Moreover, a special phenomenon named asymmetry oscillation-like convergence (AOLC) in the changing process of CE is uncovered in the source decoder and channel decoder of this system meanwhile, and the proposed joint CE criterion can reduce the impact from the AOLC phenomenon. Comparing to the counterparts, the results show that the joint CE criterion can perform well in the decoding complexity and decoding latency in the low–moderate signal-to-noise ratio (SNR) region and achieve performance improvement in the high SNR region with appropriate parameters, which also demonstrates that this system with joint CE is a low-latency and low-power system. [ABSTRACT FROM AUTHOR]

Heart sound signals reflect valuable information about heart condition. Previous studies have suggested that the information contained in single-channel heart sound signals can be used to detect coronary artery disease (CAD). But accuracy based on single-channel heart sound signal is not satisfactory. This paper proposed a method based on multi-domain feature fusion of multi-channel heart sound signals, in which entropy features and cross entropy features are also included. A total of 36 subjects enrolled in the data collection, including 21 CAD patients and 15 non-CAD subjects. For each subject, five-channel heart sound signals were recorded synchronously for 5 min. After data segmentation and quality evaluation, 553 samples were left in the CAD group and 438 samples in the non-CAD group. The time-domain, frequency-domain, entropy, and cross entropy features were extracted. After feature selection, the optimal feature set was fed into the support vector machine for classification. The results showed that from single-channel to multi-channel, the classification accuracy has increased from 78.75% to 86.70%. After adding entropy features and cross entropy features, the classification accuracy continued to increase to 90.92%. The study indicated that the method based on multi-domain feature fusion of multi-channel heart sound signals could provide more information for CAD detection, and entropy features and cross entropy features played an important role in it. [ABSTRACT FROM AUTHOR]

In recent years, on the basis of drawing lessons from traditional neural network models, people have been paying more and more attention to the design of neural network architectures for processing graph structure data, which are called graph neural networks (GNN). GCN, namely, graph convolution networks, are neural network models in GNN. GCN extends the convolution operation from traditional data (such as images) to graph data, and it is essentially a feature extractor, which aggregates the features of neighborhood nodes into those of target nodes. In the process of aggregating features, GCN uses the Laplacian matrix to assign different importance to the nodes in the neighborhood of the target nodes. Since graph-structured data are inherently non-Euclidean, we seek to use a non-Euclidean mathematical tool, namely, Riemannian geometry, to analyze graphs (networks). In this paper, we present a novel model for semi-supervised learning called the Ricci curvature-based graph convolutional neural network, i.e., RCGCN. The aggregation pattern of RCGCN is inspired by that of GCN. We regard the network as a discrete manifold, and then use Ricci curvature to assign different importance to the nodes within the neighborhood of the target nodes. Ricci curvature is related to the optimal transport distance, which can well reflect the geometric structure of the underlying space of the network. The node importance given by Ricci curvature can better reflect the relationships between the target node and the nodes in the neighborhood. The proposed model scales linearly with the number of edges in the network. Experiments demonstrated that RCGCN achieves a significant performance gain over baseline methods on benchmark datasets. [ABSTRACT FROM AUTHOR]

We address the problem of unsupervised anomaly detection for multivariate data. Traditional machine learning based anomaly detection algorithms rely on specific assumptions of normal patterns and fail to model complex feature interactions and relations. Recently, existing deep learning based methods are promising for extracting representations from complex features. These methods train an auxiliary task, e.g., reconstruction and prediction, on normal samples. They further assume that anomalies fail to perform well on the auxiliary task since they are never trained during the model optimization. However, the assumption does not always hold in practice. Deep models may also perform the auxiliary task well on anomalous samples, leading to the failure detection of anomalies. To effectively detect anomalies for multivariate data, this paper introduces a teacher-student distillation based framework Distillated Teacher-Student Network Ensemble (DTSNE). The paradigm of the teacher-student distillation is able to deal with high-dimensional complex features. In addition, an ensemble of student networks provides a better capability to avoid generalizing the auxiliary task performance on anomalous samples. To validate the effectiveness of our model, we conduct extensive experiments on real-world datasets. Experimental results show superior performance of DTSNE over competing methods. Analysis and discussion towards the behavior of our model are also provided in the experiment section. [ABSTRACT FROM AUTHOR]