Search

With the goal of developing an accurate and fast image reconstruction algorithm for ultrasound computed tomography, we combine elements of model- and data-driven approaches and propose a learned method which addresses the disadvantages of both approaches. We design a deep neural network which accounts for a nonlinear forward operator and primal-dual algorithm by its inherent network architecture. The network is trained end-to-end, with ultrasound pressure field data as input to get directly an optimized reconstruction of speed of sound and attenuation images. The training and test data are based on a set of Optical and Acoustic Breast Phantom Database, where we use the image as ground truth and simulate pressure field data according to our forward model. Extensive experiments show that our method achieves significant improvements over state-of-the-art reconstruction methods in this field. Experiments show that the proposed algorithm improves the measures structural similarity measure (SSIM) from 0.74 to 0.95 and root mean squared error (RMSE) from 0.13 to 0.09 on average concerning the speed of sound reconstruction, while it improves the SSIM from 0.60 to 0.94 and RMSE from 0.24 to 0.10 on average in attenuation reconstruction.

This article investigates a robust guaranteed cost finite-time control for coupled neural networks with parametric uncertainties. The parameter uncertainties are assumed to be time-varying norm bounded, which appears on the system state and input matrices. The robust guaranteed cost control laws presented in this article include both continuous feedback controllers and intermittent feedback controllers, which were rarely found in the literature. The proposed guaranteed cost finite-time control is designed in terms of a set of linear-matrix inequalities (LMIs) to steer the coupled neural networks to achieve finite-time synchronization with an upper bound of a guaranteed cost function. Furthermore, open-loop optimization problems are formulated to minimize the upper bound of the quadratic cost function and convergence time, it can obtain the optimal guaranteed cost periodically intermittent and continuous feedback control parameters. Finally, the proposed guaranteed cost periodically intermittent and continuous feedback control schemes are verified by simulations.

To improve the energy efficiency of a direct expansion air conditioning (DX A/C) system while guaranteeing occupancy comfort, a hierarchical controller for a DX A/C system with uncertain parameters is proposed. The control strategy consists of an open loop optimization controller and a closed-loop guaranteed cost periodically intermittent-switch controller (GCPISC). The error dynamics system of the closed-loop control is modelled based on the GCPISC principle. The difference, compared to the previous DX A/C system control methods, is that the controller designed in this paper performs control at discrete times. For the ease of designing the controller, a series of matrix inequalities are derived to be the sufficient conditions of the lower-layer closed-loop GCPISC controller. In this way, the DX A/C system output is derived to follow the optimal references obtained through the upper-layer open loop controller in exponential time, and the energy efficiency of the system is improved. Moreover, a static optimization problem is addressed for obtaining an optimal GCPISC law to ensure a minimum upper bound on the DX A/C system performance considering energy efficiency and output tracking error. The advantages of the designed hierarchical controller for a DXA/C system with uncertain parameters are demonstrated through some simulation results.

Multi-Label Learning (MLL) is a classification task in which each instance may be associated with two or more class labels simultaneously. With the arrival of information explosion era, MLL is confronted with the curse of high-dimensional data. Feature selection is an effective technique that has been used to reduce dimensionality. Previous Multi-Label Feature Selection (MLFS) approaches only lay emphasis on the fitting relations between the feature space and the corresponding label space while neglecting the instance correlations. To remedy the deficiency, we present a novel MLFS method by Exploring Instance Correlations with Local Discriminant model (EICLD). More specifically, we first construct a local set for each instance, and use a local discriminant model for each local set to explore instance correlations. Then, we further integrate the local models of all instances to guarantee the global instance correlations. Finally, l2,1-norm is introduced into loss function and the regularization of feature weight matrix respectively to facilitate feature selection process. An optimization algorithm is designed for handling the proposed objective function. The paper is compared with seven representative approaches on twelve data sets. The results validate the superiority of the proposed approach.

In multi-label learning, feature selection is an essential preprocessing module, which can be exploited a more compact and precise representation of instances. Most of existing multi-label feature selection methods are either converted into multiple single-label feature selection methods or directly utilize half-baked label information, thus it is difficult for them to obtain a discriminative feature subset across multiple labels. To tackle this problem, we propose multi-label feature selection with local discriminant model and label correlations. First, for each instance, a local clique comprising this instance and its neighboring instances is constructed, and a local discriminant model for each local clique is integrated globally to evaluate the clustering performance of all instances. Second, in terms of clustering results, we explore high-order label correlations to reduce the impact of half-baked label information. Finally, we combine l2,1-norm regularization to design the objective function to achieve multi-label feature selection. Comprehensive experiments are conducted on twelve real-world multi-label data sets, and results demonstrate the effectiveness of the proposed method in comparison with several representative methods.

The Laplacian eigenmaps (LE) is one of the most commonly used nonlinear dimensionality reduction methods and aims to find a low-dimensional representation to preserve the topological relationship between sample points in the original data. However, the l2-norm based loss function makes LE unable to preserve the relationship in many cases. Additionally, the topological relationship does not represent the real intrinsic structure of data. For example, the overemphasis of the topological relationship by LE easily breaks the manifold structure into multiple local areas in the embedding space, which makes the spectral clustering analysis of multi-manifold data more difficult to carry out. To solve this problem, we propose the soft adaptive loss based LE (SALE). With the soft adaptive loss, SALE can adaptively emphasize the topological relationship between sample points and the clustering structure of data. The model is tested and validated on UCI, face and gene expression data sets, and compared with some state-of-the-art models. The experimental results show that the method is robust to noise.

In the literature, a lot of work focused on studying intermittent control problems via feedback control strategy. No study on the intermittent control problems via sliding mode control method has been reported so far. This paper studies the problem of fast synchronization between two complex dynamical networks via aperiodically intermittent control and sliding mode control. In order to achieve fast synchronization of complex dynamical networks by using aperiodically intermittent sliding mode controller, new differential inequalities are derived firstly. After that, some sufficient finite-time synchronization criteria and finite-time achieving slide mode surface are obtained based on finite-time stability theory, aperiodically intermittent sliding mode control technique and constructing Lyapunov function. Finally, an example is provided to verify the effectiveness of the proposed theoretical methods.

In this paper, the finite-time synchronization problem of complex dynamic networks with time delay is studied via aperiodically intermittent control. By compared with the existed results concerning aperiodically intermittent control, some new results are obtained to guarantee the synchronization of networks in a finite time. Especially, a new lemma is proposed to reduce the convergence time. In addition, based on aperiodically intermittent control scheme, the essential condition ensuring finite-time synchronization of dynamic networks is also obtained, and the convergence time is closely related to the topological structure of networks and the maximum ratio of the rest width to the aperiodic time span. Finally, a numerical example is provided to verify the validness of the proposed theoretical results.

Deep neural networks have proven to excel classical medical image reconstruction techniques. Some networks are based on fully connected (FC) layers to achieve domain transformation such as from the data acquisition domain to the image domain. However, FC layers result in huge numbers of parameters which take a lot of GPU memory. Hence, they do not scale well, and the overall performance is limited. For ultrasound computed tomography (USCT) application, we propose a memory-efficient convolutional network that reconstructs images from the frequency domain to image domain with much less parameters compared with multilayer perceptron, by using data-driven learning. Extensive experiments demonstrate that our method achieves high reconstruction quality. It improves the structural similarity measure (SSIM) from 0.73 to 0.99 when compared with state-of-the-art reconstruction methods in this field while reduces 2/3 parameters when compared with deep neural network with FC layers to reconstruct images from frequency domain to image domain.

Ultrasound transmission tomography promises a high potential and novel imaging method for early breast cancer diagnosis; it can quantitatively characterize tissues or materials by the attenuation and speed of sound (SoS). Reconstruction of ultrasound transmission tomography is an inverse problem that can be solved iteratively based on a paraxial approximation of the Helmholtz equation as forward model, which is highly non-linear and time-consuming. In order to address these problems and reconstruct desired images, we design a dual domain network architecture for ultrasound transmission tomography reconstruction. It can enhance the information of measurement domain and directly reconstruct from pressure field measurements without using any initialization of reconstruction and fully connected layer. We train the network on simulated ImageNet data and transfer it for ultrasound transmission tomography images to avoid overfitting when the amount of ultrasound transmission tomography images is limited. Our experimental results demonstrate that a dual domain network produces significant improvements over state-of-the-art methods. It improves the measured structural similarity measure (SSIM) from 0.54 to 0.90 and normalized root mean squared error (nRMSE) from 0.49 to 0.01 on average concerning the SoS reconstruction, and from 0.46 to 0.98 for SSIM, from 353 to 0.03 for nRMSE on average concerning the attenuation reconstruction.

Network security has become an important issue in our work and life. Hackers' attack mode has been upgraded from normal attack to APT( Advanced Persistent Threat, APT) attack. The key of APT attack chain is the penetration and intrusion of active directory, which can not be completely detected via the traditional IDS and antivirus software. Further more, lack of security protection of existing solutions for domain control aggravates this problem. Although researchers have proposed methods for domain attack detection, many of them have not yet been converted into effective market-oriented products. In this paper, we analyzes the common domain intrusion methods, various domain related attack behavior characteristics were extracted from ATT&CK matrix (Advanced tactics, techniques, and common knowledge) for analysis and simulation test. Based on analyzing the log file generated by the attack, the domain attack detection rules are established and input into the analysis engine. Finally, the available domain intrusion detection system is designed and implemented. Experimental results show that the network attack detection method based on the analysis of domain attack behavior can analyze the log file in real time and effectively detect the malicious intrusion behavior of hackers , which could facilitate managers find and eliminate network security threats immediately.

Ultrasound Computed Tomography (USCT) is a new high resolution imaging method with high potential for breast cancer detection. USCT image reconstruction requires the iterative solution of a wave equation and its adjoint, being very time-consuming. To address this time consumption problem, a convolutional neural network with frequency pressure field as input is presented, outperforming classical iterative optimization strategies to solve the inverse problem. The Optical and Acoustic Breast Phantom Database (OA-Breast) is used to generate simulated USCT data using paraxial approximation forward model both for training and testing. Extensive experiments demonstrate that our neural network outperforms the classical strategies both in reconstruction time and quality.

Raw point cloud data are often noisy, superfluous, and with topological defects, such as holes. These issues cause inaccurate geometric representation in 3D reconstruction. As a result, surface reconstruction from point cloud data is a highly challenging problem. In this paper, we address the aforementioned issues by taking advantage of the embedded information of segmentation, skeletonization, and user guidance. First, we pre-process the point cloud data with three steps, relocating each point, upsampling the point data, and optimizing normals to enhance the features and geometric details; second, a segmentation method converts the input cloud into separate parts; finally, we construct curve skeletons for each part and guide the surface reconstruction with minimal user interaction, where the parts of refined smooth shapes are fused to generate the final results. The comparison studies confirmed that the proposed method is able to produce state-of-the-art results in terms of preserving sharp features, handling missing data, and requiring minimal user intervention.

Determining a systematic whole-building retrofit plan for envelope components and indoor appliances to achieve targets such as cost savings and policy compliance is a challenging task. To be specific, the systematic whole-building retrofit problem, when solved by an optimization approach, is highly complicated. It is sometimes even impossible to find a solution with given computation resources and algorithms. In addition, a costly comprehensive bottom-up audit is required to establish the parameters of the problem. This study presents two models to reduce the complexity of the systematic whole-building retrofit optimization problem. Firstly, the proposed models use the grouping concept to optimize the retrofit of subsystems in a building instead of individual components/appliances, which reduces the dimension of the problem effectively. Secondly, the models use so-called ‘notch test’ data, which are sampled and verified savings of an intervention, to eliminate the need for bottom-up energy audits. This further simplifies the retrofit optimization problem and reduces the retrofit cost. The models are based on our previous work and aim at energy savings maximization and payback period minimization, considering the green building policy and tax incentive initiatives. A case study shows that about 2530 MWh energy savings and an A rating from the energy performance certificate standard can be obtained with a payback period of 59 months, which verifies the feasibility and effectiveness of the models proposed.

Nowadays, multi-label learning is ubiquitous in practical applications, in which multi-label data is always confronted with the curse of high-dimensional features. Feature selection has been shown to effectively improve learning performance by selecting discriminative features. Conventional multi-label feature selection only focuses on associating input features with corresponding labels while neglecting the potential structural information, i.e., instance correlations and label correlations. To tackle this problem, we propose manifold learning with structured subspace for multi-label feature selection. Specifically, we first uncover a latent subspace for a more compact and accurate data representation, and take advantage of the subspace to explore the correlations among instances. Then, we explore label correlations in manifold learning to guarantee the global and local structural consistency of labels. Besides, l 2 , 1 -norm is introduced into loss function and sparse regularization to facilitate feature selection process. A detail optimization algorithm is presented to solve the objective function of the proposed method. Extensive experiments on real-world data show the superiority of the proposed method under various metrics.

The segmentation of a point cloud is one of the key technologies for three-dimensional reconstruction, and the segmentation from three-dimensional views can facilitate reverse engineering. In this paper, we propose a self-adaptive segmentation algorithm, which can address challenges related to the region-growing algorithm, such as inconsistent or excessive segmentation. Our algorithm consists of two main steps: automatic selection of seed points according to extracted features and segmentation of the points using an improved region-growing algorithm. The benefits of our approach are the ability to select seed points without user intervention and the reduction of the influence of noise. We demonstrate the robustness and effectiveness of our algorithm on different point cloud models and the results show that the segmentation accuracy rate achieves 96%.

Like single-label learning, multi-label learning also suffers from the curse of dimensionality. Due to the existence of high-dimensional data, feature selection as a preprocessing tool always plays a key position in the multi-label learning. Although a variety of multi-label feature selection methods have been developed, they neglect to consider the redundancy of features, thereby degrading learning performance. To address this problem, we present a novel multi-label feature selection approach with uncorrelated regression and adaptive spectral graph. Specifically, we first construct a manifold framework with uncorrelated regression model to hunt for uncorrelated yet discriminative features, which also utilize the low-dimensional representation based on feature space to fit the label distribution. Then, a spectral graph term based on information entropy is incorporated into the manifold framework, so as to ensure the local geometric structure of data in subsequent learning process. Following the above principles, we design an objective function to achieve multi-label feature selection, and propose a detailed optimization method. Comprehensive experiments are conducted on multiple public multi-label data sets, the results show that our method is superior to other compared methods.

To reduce the fluctuation of renewable energy (RE) supply and improve the economic efficiency of the power grid, the energy local area network (ELAN), which is a subnetwork of the energy internet (EI), plays an important role in specific regions. Electric vehicles (EVs), as virtual energy storage (VES) in ELANs, are helpful to decrease the fluctuations of RE supply. However, how to use EVs in ELANs is a complex issue, considering the uncertainties of EVs’ charging demand, the forecast data errors of RE sources, etc. In this paper, a typical ELAN structure is established, taking into account RE sources, load response system, and a distributed energy storage (DES) system including EVs. A two-step optimization framework for ELAN scheduling problem is proposed. A global optimization model based on forecast data is built to maximize the income of ELAN, and an online local optimization model is introduced to minimize the correction cost utilizing prior knowledge. Finally, the proposed two-step optimization framework is applied to a series of real-world ELAN scheduling problems. The results show that DES system with EVs can reduce the volatility of RE supply evidently, and the proposed method is able to maximize the income of the ELAN efficiently.

The rise of skeletal skinning technology has provided great convenience for animators. At the same time, it improves the efficiency of animation production. However, the deformation resulting from this technology suffers from some undesirable effects, which require manual improvement. In this paper, we propose an approach addressing the problem of creating believable mesh-based skin deformation. In this approach, the skin is first deformed with a classic linear blend skinning approach, which usually lead to artifacts like the candy-wrapper effect or volume loss. Then we enforce the geometric constraints which displace the positions of the vertices to mimic the behavior of the skin and achieve effects like volume preservation. At last, we adopt the finite element method to handle large deformed elements which could accelerate the system’s convergence rate. This approach is easy to implement and has a high skinning efficiency without affecting the simulating effect.

In the process of auto-generating test-paper, the category and the difficulty of the title plays a key role in the quality of generating test-paper. It will produce low quality questions and hard to popularize when used the methods of artificial generating test-paper and random generating test-paper, because considering less on the knowledge point classification and difficulty in the subject. To improve the quality of auto-generating test-paper, this paper takes the evaluation data of ACM Online Judge system as the research object. After normalization, (1) we can get three different results by the K-means clustering analysis based on the temporal and spatial characteristics of time variance and average time; (2) On the basis of clustering, the difficulty index of each topic of all the categories is calculated by using the number of submissions and the number of submissions to solve the problem. The ratio of the two is proportional to the difficulty of the problem. In this paper, the ratio of the two to determine the degree of difficulty index; (3) The Gaussian stochastic process is used to extract numbers of questions of each knowledge point, and calculated the difficulty index which were extracted to make sure they are within range to complete the auto-generating test-paper. In the experiment, we try to train and test the automatic test paper model by the number of professional problems (about 50000 data) in the C language test question of the university OJ system. The average difficulty index of the test paper was 0.4663, which meet the requirements, and the difficulty index of the title fit in with the normal distribution. Compared with the traditional generating test-paper method, the automatic test paper model is based on the difficulty and discrimination of the subject, and it can evaluate the level of tester scientifically. The experimental results show that the proposed automatic test model is simple and effective.

Raw data of point cloud is often noisy and with\ud topological defects (such as holes), which cause problems including faulty connection and inaccurate structure. As\ud a result, the surface reconstruction of point cloud data is\ud a highly challenging problem. This work proposes a novel\ud method, which improves the surface quality compared with\ud existing methods. Our method combines both the local detailed\ud features and the global topological information during the\ud reconstruction process. To facilitate the feature refinement,\ud we first pre-process the point cloud data by relocating each\ud point, upsampling the point data, and optimizing normals to\ud enhance the features and geometric details. We then identify\ud the topological information by segmenting the geometry and\ud constructing curve skeletons for each part and guide the surface reconstruction with the skeletons by minimal user interaction. We demonstrate the effectiveness of our methods with various examples, where our reconstruction can fill out missing data and preserve sharp features.

Phosphorylation is an important part of post-translational modifications of proteins, and is essential for many biological activities. Phosphorylation and dephosphorylation can regulate signal transduction, gene expression, and cell cycle regulation in many cellular processes. Phosphorylation is extremely important for both basic research and drug discovery to rapidly and correctly identify the attributes of a new protein phosphorylation sites. Moreover, abnormal phosphorylation can be used as a key medical feature related to a disease in some cases. The using of computational methods could improve the accuracy of detection of phosphorylation sites, which can provide predictive guidance for the prevention of the occurrence and/or the best course of treatment for certain diseases. Furthermore, this approach can effectively reduce the costs of biological experiments. In this study, a flexible neural tree (FNT), particle swarm optimization, and support vector machine algorithms were used to classify data with secondary encoding according to the physical and chemical properties of amino acids for feature extraction. Comparison of the classification results obtained from the three classifiers showed that the classification of the FNT was the best. The three classifiers were then integrated in the form of a minority subordinate to the majority vote to obtain the results. The performance of the integrated model showed improvement in sensitivity (87.41%), specificity (87.60%), and accuracy (87.50%).

To tackle with the blindness of random questions choosing for exercise and test of the on-line learning system, this paper clusters questions exploiting various feature subsets and parameters via K-means. For the test data of ACM Online Judge system, the features of temporal fluctuations mean of time consumption and repeat submission rate are used to make the question categorization and automatic recommendation come true. The experimental results suggest that the proposed method is simple but effective, and by which an on-line test platform can realize functions such as individuation teaching, intelligently questions choosing, teaching instruction, automatically paper constructing and paper difficult prediction.

Performing microarray expression data classification can improve the accuracy of a cancer diagnosis. The varying technique including Support Vector Machines (SVMs), Neuro-Fuzzy models (NF), K-Nearest Neighbor (KNN), Neural Network (NN), and etc. have been applied to analyze microarray expression data. In this investigation, a novel complex network classifier is proposed to do such thing. To build the complex network classifier, we tried a hybrid method based on the Particle Swarm Optimization algorithm (PSO) and Genetic Programming (GP), of which GP aims at finding an optimal structure and PSO accomplishes the fine tuning of the parameters encoded in the proposed classifier. The experimental results conducted on Leukemia and Colon data sets are comparable to the state-of-the-art outcomes.

This paper presents a novel approach for bas-relief generation and synthesis. In contrast to previous methods, we divide this problem into two parts: the selection of the best view and arrangement of the relief layout. Taking these into account, we incorporate the visual saliency and photographic composition rules into the bas-relief generation. Additionally, a nonlinear compression function is used to compress the models, and finally, we implement surface parameterization by directly manipulating the mesh triangles to generate curved surface bas-relief. We validate our approach through a variety of models. The results indicate that the proposed approach is effective to adapt different types of target surface with topology unchanged. Comparing with conventional methods, our approach is able to effectively produce bas-relief with a reasonable layout and distinct details.

Medical Guidelines pay an important role in medical decision making systems. Medical guidelines are usually involved with event-related actions or procedure. However, little research has been done on how event-related medical guidelines can be converted into its semantic representation such as RDF/OWL data. This paper proposes an approach of automatic generation of semantic data for event-related medical guidelines. This generation is achieved by using the logic programming language Prolog with the support of medical ontologies such as SNOMED CT. We will report the experiments with the automatic generation of the semantic data for event-related Chinese medical guidelines, and show the relevant results.

Accurate classification on protein subcellular localization plays an important role in Bioinformatics. An increasingly evidences demonstrate that a variety of classification methods have been employed in this field. This research adopts feature fusion method to extract the information of the protein subcellular. Several types of features are employed in this protein coding method, which include amino acid index distribution, the stereo-chemical properties of amino acids and the information for local sequence of amino acids. On base of this feature combination method, flexible neutral tree (FNT) is employed to predict multiplex protein subcellular locations. The overall accuracy rate of using flexible neutral tree as prediction algorithm may reach a better result.

This paper proposes a novel feature fusion method for the protein subcellular multiple-site localization prediction. Several types of features are employed in this novel protein coding method. The first one is the composition of amino acids. The second is pseudo amino acid composition, which mainly extract the location information of each amino acid residues in protein sequence. Lastly, the information for local sequence of amino acids is taken into consideration in this research. Generally, k nearest neighbor, supporting vector machine and other methods, has been used in the field of protein subcellular localization prediction. In our research, the multi-label k nearest neighbor algorithm has been employed in the classification model. The overall accuracy rate may reach 66.7304% in Gnos-mploc dataset.

Clinical practice guidelines aim to help doctors and patients to improve the quality of care, reduce unjustified practice variations and reduce healthcare costs. The study of medical events is of special significance in medicine. However, in most existing medical guidelines, there is a lack of effective methods and tools to describe medical events. This paper analyzes the medical guidelines and the events in those guidelines, discusses the definition of semantic event and the theoretical model of event proposed by Vendler. On this basis, the paper introduces the importance as an additional dimension in the definition on events. Events of guidelines can be extracted according to this definition. We convert those extracted events using the XSLT to generate their RDF semantic data. The generated semantic data are mapped with not only their relevance of the well-known medical ontology such as SNOMED CT, but also used in the system SeSRUA, a semantically-enabled system for rational use of antibiotics. The experiments show that it can promote the rational use of drugs in the development of information technology and knowledge management.