Search

Computational drug design relies on the calculation of binding strength between two biological counterparts especially a chemical compound, i.e., a ligand, and a protein. Predicting the affinity of protein-ligand binding with reasonable accuracy is crucial for drug discovery, and enables the optimization of compounds to achieve better interaction with their target protein. In this paper, we propose a data-driven framework named DeepAtom to accurately predict the protein-ligand binding affinity. With 3D Convolutional Neural Network (3D-CNN) architecture, DeepAtom could automatically extract binding related atomic interaction patterns from the voxelized complex structure. Compared with the other CNN based approaches, our light-weight model design effectively improves the model representational capacity, even with the limited available training data. We carried out validation experiments on the PDBbind v.2016 benchmark and the independent Astex Diverse Set. We demonstrate that the less feature engineering dependent DeepAtom approach consistently outperforms the other baseline scoring methods. We also compile and propose a new benchmark dataset to further improve the model performances. With the new dataset as training input, DeepAtom achieves Pearson's R=0.83 and RMSE=1.23 pK units on the PDBbind v.2016 core set. The promising results demonstrate that DeepAtom models can be potentially adopted in computational drug development protocols such as molecular docking and virtual screening.

With the rapid growth of electronic chip’s performance, electric signal limits chip I/O in power consumption, reachability, and signal quality. In this letter, we propose a new chip-let architecture for chip-level optical interconnect. An optic I/O chip-let including an ultra-small optic transceiver and electronic components is implemented. Our analysis shows that the optical interconnect based on this architecture can achieve 1/3 power consumption and 1/2 area compared with traditional board-level optical interconnect in Ethernet NIC application. By adopting the architecture we proposed, the optic I/O chip-let can support any payload IC such as CPU, GPU, switch to have optic I/O.

Power grounding is a key aspectofensuringsafety in power facilities.However, the corrosion of grounding materials can cause accidentsduringpower facility operation. Therefore, monitoring the corrosion status of grounding materials can eliminate hidden risks in the grounding network and ensure safe poweroperation. In this paper, electromagnetic ultrasonic thickness measurement technology was used to develop an online corrosion monitoring system for grounding materials via the installation of electromagnetic ultrasonic measurement probes on in-service power grounding materials. The results from a substation grounding networkdemonstrate that the online corrosion monitoring system can obtain more precise grounding corrosion data and has more extensive application prospects compared with other monitoring methods.

PurposeWith the increasing abundance of network resources and big data, multi-modal information search (MMIS) has been paid more and more attention, but the research results of MMIS are relatively few. This paper attempts to put forward constructive suggestions for the design of multi-modal information system, so that the system can have a better user experience, help users improve the efficiency of obtaining information and optimize the information service mode.Design/methodology/approachA research model of influencing factors is established by using the TAM (technology acceptance model) theory. The influencing factors of users' multi-modal information search behavior (MMISB) are analyzed by using questionnaire, experiment and the structural equation model. On the basis of this, some suggestions are put forward to build the multi-modal search (MMS) system and improve the efficiency of MMIS.FindingsThe research shows that users' MMISB is directly related to their search intention, and the search intention can influence users' cognition of the usefulness and ease of MMIS through their own information search ability and system characteristics. The user's MMIS ability is affected by the demand expression ability and retrieval ability cognition; the user's cognition of system characteristics is affected by the system function and information quality. This shows that the user's MMISB is closely related to the user's cognitive situation, but due to the author's limited time and research ability, only the questionnaire survey method cannot be used to in-depth research and explore the influencing factors of MMIS. Therefore, in the future research, we should combine the interview method to further track the user's emotional factors and scene factors.Originality/valueFor the first time, TAM theory is combined with cross-modal retrieval behavior and the paper explores the influencing factors and evaluation indexes of users' MMISB. The second, the questionnaire was compiled to investigate the influencing factors of the MMISB of the university group, and the reliability analysis, validity analysis, correlation analysis and structural equation model analysis of the survey data are carried out . The survey data and analysis results are original, which can provide a theoretical basis for improving the service level of MMIS.

We propose a mathematical model based on probability theory to optimize COVID‐19 testing by a multistep batch testing approach with variable batch sizes. This model and simulation tool dramatically increase the efficiency and efficacy of the tests in a large population at a low cost, particularly when the infection rate is low. The proposed method combines statistical modeling with numerical methods to solve nonlinear equations and obtain optimal batch sizes at each step of tests, with the flexibility to incorporate geographic and demographic information. In theory, this method substantially improves the false positive rate and positive predictive value as well. We also conducted a Monte Carlo simulation to verify this theory. Our simulation results show that our method significantly reduces the false negative rate. More accurate assessment can be made if the dilution effect or other practical factors are taken into consideration. The proposed method will be particularly useful for the early detection of infectious diseases and prevention of future pandemics. The proposed work will have broader impacts on medical testing for contagious diseases in general.

In the web service marketplace, component-based economy has been proposed for describing participants' behavioral patterns. Composite service networks combine multiple composite services required by various service consumers. With each composite service as a product, web services comprise heterogeneous products. In this study, the pricing behavior of networked individual service providers is investigated. With the objective of service survival or high profitability, service providers compete both on the single-service and service-network levels. Using examples, several mild assumptions are formulated and analyzed. Then, a bi-objective optimization model is proposed based on these assumptions, which attempts to maintain a reasonable effectiveness-fairness trade-off from the individual service providers' perspective. The NP-completeness of the single-objective version is demonstrated by transforming the problem into a subset sum problem, which highlights the challenge of obtaining a pareto set for the bi-objective model. Finally, to validate the proposed model, numerical experimentation and case study are conducted, and both the bi-objective and many-objective versions of the problem are discussed.

In this paper, we exploit the idea of virtual ESPRIT (VESPA) to develop a multi-baseline VESPA (MB-VESPA) approach for direction finding. Specially, we define several cumulant matrices to provide ambiguous direction estimates under different baselines. Fine and unambiguous estimation is then obtained by a simple refinement step. Two refine approaches, termed as successive baseline approach and coprime baseline approach, are subsequently introduced. MB-VESPA shares all the advantages of the VESPA. It is simple, closed-form, search-free, and is applicable to irregularly linear array. In addition, it is free of the impact on the sensor gain uncertainties.

In this paper, we focus on the task query-based video localization, i.e., localizing a query in a long and untrimmed video. The prevailing solutions for this problem can be grouped into two categories: i) Top-down approach: It pre-cuts the video into a set of moment candidates, then it does classification and regression for each candidate; ii) Bottom-up approach: It injects the whole query content into each video frame, then it predicts the probabilities of each frame as a ground truth segment boundary (i.e., start or end). Both two frameworks have respective shortcomings: the top-down models suffer from heavy computations and they are sensitive to the heuristic rules, while the performance of bottom-up models is behind the performance of top-down counterpart thus far. However, we argue that the performance of bottom-up framework is severely underestimated by current unreasonable designs, including both the backbone and head network. To this end, we design a novel bottom-up model: Graph-FPN with Dense Predictions (GDP). For the backbone, GDP firstly generates a frame feature pyramid to capture multi-level semantics, then it utilizes graph convolution to encode the plentiful scene relationships, which incidentally mitigates the semantic gaps in the multi-scale feature pyramid. For the head network, GDP regards all frames falling in the ground truth segment as the foreground, and each foreground frame regresses the unique distances from its location to bi-directional boundaries. Extensive experiments on two challenging query-based video localization tasks (natural language video localization and video relocalization), involving four challenging benchmarks (TACoS, Charades-STA, ActivityNet Captions, and Activity-VRL), have shown that GDP surpasses the state-of-the-art top-down models.

Taking an answer and its context as input, sequence-to-sequence models have made considerable progress on question generation. However, we observe that these approaches often generate wrong question words or keywords and copy answer-irrelevant words from the input. We believe that lacking global question semantics and exploiting answer position-awareness not well are the key root causes. In this paper, we propose a neural question generation model with two general modules: sentence-level semantic matching and answer position inferring. Further, we enhance the initial state of the decoder by leveraging the answer-aware gated fusion mechanism. Experimental results demonstrate that our model outperforms the state-of-the-art (SOTA) models on SQuAD and MARCO datasets. Owing to its generality, our work also improves the existing models significantly.

Scholarships are a reflection of academic achievement for college students. The traditional scholarship assignment is strictly based on final grades and cannot recognize students whose performance trend improves or declines during the semester. This paper develops the Trajectory Mining on Clustering for Scholarship Assignment and Academic Warning (TMS) approach to identify the factors that affect the academic achievement of college students and to provide decision support to help low-performing students attain better performance. Specifically, we first conduct feature engineering to generate a set of features to characterize the lifestyles patterns, learning patterns, and Internet usage patterns of students. We then apply the objective and subjective combined weighted k-means (Wosk-means) algorithm to perform clustering analysis to identify the characteristics of different student groups. Considering the difficulty in obtaining the real global positioning system (GPS) records of students, we apply manually generated spatiotemporal trajectories data to quantify the direction of trajectory deviation with the assistance of the PrefixSpan algorithm to identify low-performing students. The experimental results show that the silhouette coefficient and Calinski-Harabasz index of the Wosk-means algorithm are both approximately 1.5 times to that of the best baseline algorithm, and the sum of the squared error of the Wosk-means algorithm is only the half of the best baseline algorithm.

A long-term goal of Artificial Intelligence (AI) is to provide machines with the capability of understanding natural language. Understanding natural language may be referred as the system must produce a correct response to the received input order. This response can be a robot move, an answer to a question, etc. One way to achieve this goal is semantic parsing. It parses utterances into semantic representations called logical form, a representation of many important linguistic phenomena that can be understood by machines. Semantic parsing is a fundamental problem in natural language understanding area. In recent years, researchers have made tremendous progress in this field. In this paper, we review recent algorithms for semantic parsing including both conventional machine learning approaches and deep learning approaches. We first give an overview of a semantic parsing system, then we summary a general way to do semantic parsing in statistical learning. With the rise of deep learning, we will pay more attention on the deep learning based semantic parsing, especially for the application of Knowledge Base Question Answering (KBQA). At last, we survey several benchmarks for KBQA.

There are generally many redundant and irrelevant features in high-dimensional datasets, which leads to the decline of classification performance and the extension of execution time. To tackle this problem, feature selection techniques are used to screen out redundant and irrelevant features. The artificial bee colony (ABC) algorithm is a popular meta-heuristic algorithm with high exploration and low exploitation capacities. To balance between both capacities of the ABC algorithm, a novel ABC framework is proposed in this paper. Specifically, the solutions are first updated by the process of employing bees to retain the original exploration ability, so that the algorithm can explore the solution space extensively. Then, the solutions are modified by the updating mechanism of an algorithm with strong exploitation ability in the onlooker bee phase. Finally, we remove the scout bee phase from the framework, which can not only reduce the exploration ability but also speed up the algorithm. In order to verify our idea, the operators of the grey wolf optimization (GWO) algorithm and whale optimization algorithm (WOA) are introduced into the framework to enhance the exploitation capability of onlooker bees, named BABCGWO and BABCWOA, respectively. It has been found that these two algorithms are superior to four state-of-the-art feature selection algorithms using 12 high-dimensional datasets, in terms of the classification error rate, size of feature subset and execution speed.

The accurate prediction of protein-ligand binding affinity is critical for the success of computer-aided drug discovery. However, the accuracy of current scoring functions is usually unsatisfactory due to their rough approximation or sometimes even omittance of many factors involved in protein-ligand binding. For instance, the intrinsic dynamic of the protein-ligand binding state is usually disregarded in scoring function because these rapid binding affinity prediction approaches are only based on a representative complex structure of the protein and ligand in the binding state. That is, the dynamic protein-ligand binding complex ensembles are simplified as a static snapshot in calculation. In this study, two novel features were proposed for characterizing the dynamic properties of protein-ligand binding based on the static structure of the complex, which is expected to be a valuable complement to the current scoring functions. The two features demonstrate the geometry-shape matching between a protein and a ligand as well as the dynamic stability of protein-ligand binding. We further combined these two novel features with several classical scoring functions to develop a binary classification model called DyScore that uses the Extreme Gradient Boosting algorithm to classify compound poses as binders or non-binders. We have found that DyScore achieves state-of-the-art performance in distinguishing active and decoy ligands on both enhanced DUD dataset and external test sets with both proposed novel features showing significant contributions to the improved performance. Especially, DyScore exhibits superior performance on early recognition, a crucial requirement for success in virtual screening andde novodrug design. The standalone version of DyScore and Dyscore-MF are freely available to all at:https://github.com/YanjunLi-CS/dyscoreKey PointsTwo novel binding features were proposed for characterizing the dynamic properties of protein-ligand binding only based on a static snapshot of complex.Based on the XGBoost machine learning method, the DyScore recognition model was proposed to accurately classify compound binding poses as binders or non-binders. DyScore consistently outperforms all the state-of-the-art published models on three different metrics by a large margin.DyScore showed superior performance in early recognition with an average of 73.3% success rate for the top three ranked compounds for each protein target.The standalone version of DyScore and DyScore-MF are freely available to all at:https://github.com/YanjunLi-CS/dyscoreTOC

This paper presents a 32 Gb/s transmitter (TX) with non-return-to-zero (NRZ) modulation scheme. Fabricated in the 28nm CMOS technology, the TX incorporates a quarter-rate architecture with a three-tap feed-forward equalizer (FFE) under the consideration of timing requirement and power consumption. Key features of the TX include a low-power data-serializing path, a pulsed pass-gate 4:1 multiplexer (MUX), a pre-driver adopted sub-UI de-emphasis, a current-mode logic (CML) output driver combined common-mode feedback (CMFB), an output network using a T-coil inductor to eliminate the parasitic capacitance of ESD and pad for bandwidth expansion. The critical clock path contains sub-50fs resolution duty cycle detection/correction (DCD/DCC) and quadrature error detection/correction (QED/QEC) circuits. The TX operating at 32Gb/s in NRZ modulation, including the clock path, consumes 98 mW under 1V supply, achieving a 3.06 pJ/b energy efficiency with 0.8 V pp output swing. The TX front end core circuit occupies an area of 0.078 mm2.

In this paper, we conduct a comprehensive study on insect identification and classification. However, the existing insect datasets are made up of several categories, which is far from the demands in reality. To handle this issue, we contribute to a more challenging insect image dataset, which contains 1848 images covering different pests from 118 classes. We further conduct fine-tuning on seven deep convolution neural networks, including VGG16, ResNet50, DenseNet121, Res2Net50_26w_ 4s, SCNet50_ vld, GhostNet, and RegNet. Finally, extensive experiments of the aforementioned networks on our dataset illustrate that the last four state-of-the-art deep convolution neural networks can achieve promising performance on pest identification and classification.

Aiming at the audio event recognition problem of speech recognition, a decision fusion method based on the Transformer and Causal Dilated Convolutional Network (TCDCN) framework is proposed. This method can adjust the model sound events for a long time and capture the time correlation, and can effectively deal with the sparsity of audio data. At the same time, our dataset comes from audio clips cropped by YouTube. In order to reliably and stably identify audio topics, we extract different features and different loss function calculation methods to find the best model solution. The experimental results from different test models show that the TCDCN model proposed in this paper achieves better recognition results than the classification using neural networks and other fusion methods.

Aiming at the problems of long legal document entities and the lack of annotated data, a cascade model that integrates the characteristics of characters and words is proposed. The traditional NER is decomposed into two cascaded subtasks: the entity recognition and the attribute recognition. The model obtains the vector representation of text character-level and word-level through BERT and Self-attention, respectively. Then the BiLSTM is used to obtain the internal features of the serialized text. Subsequenly, CRF is used to select the entity's optimal tag sequence and the attributed optimal tag sequence. Finally, these two sequences are spliced to obtain the optimal marker sequence. In order to improve the utilization rate of the data, the label linearization is introduced for the data enchancement. The results show that the method is superior to traditional models, can effectively extract named entities of legal documents, and has the vital practical significance.

Aiming at the problems of poor parsing effect and poor generalization ability in Chinese position segmentation using current neural network models, this paper proposes a Chinese position segmentation method based on ALBERT-BiGRU-CRF model. The method first uses the ALBERT pre-training model to pre-train the Chinese location information to obtain the context information in all layers, enhance the semantic representation ability of the Chinese location information, and then extract the feature information of the vector through the BiLSTM model, and finally decode it through the CRF model to obtain the global Optimal labeling sequence. Experimental results show that on the basis of different numbers and regions of Chinese location information data sets, the ALBERT-BiGRU-CRF model has better word segmentation accuracy and F1 value on all test sets than the current commonly used neural network models, and the highest can be achieved. 93.91% and 93.96%. Using the ALBERT-BiGRU-CRF model to segment Chinese location information not only effectively improves the accuracy of Chinese location information analysis and polysemous word analysis, but also has better generalization capabilities.

In view of the difficulty in extracting the key area features of the image and the over-complex training in current street view image retrieval methods, this paper proposes a retrieval method that extracts and integrates multiple global features. First, the Convolutional Neural Network is used to extract the features, and then the new multi-scale pooling layer is used to output multiple global features, and the feature loss is reduced by outputting fixed-dimensional features. Finally, the final feature obtained by concatenating multiple global features is used for retrieval. Experimental results show that this method can effectively extract image features, reduce the complexity of training, and improve the accuracy of retrieval.

Urban regions are places where people live, work, consume, and entertain. In this study, we investigate the problem of learning an embedding space for regions. Studying the representations of regions can help us to better understand the patterns, structures, and dynamics of cities, support urban planning, and, ultimately, to make our cities more livable and sustainable. While some efforts have been made for learning the embeddings of regions, existing methods can be improved by incorporating locality-constrained spatial autocorrelations into an encode-decode framework. Such embedding strategy is capable of taking into account both intra-region structural information and inter-region spatial autocorrelations. To this end, we propose to learn the representations of regions via a new embedding strategy with awareness of locality-constrained spatial autocorrelations. Specifically, we first construct multi-view (i.e., distance and mobility connectivity) POI-POI networks to represent regions. In addition, we introduce two properties into region embedding: (i) spatial autocorrelations: a global similarity between regions; (ii) top-k locality: spatial autocorrelations locally and approximately reside on top k most autocorrelated regions. We propose a new encoder-decoder based formulation that preserves the two properties while remaining efficient. As an application, we exploit the learned embeddings to predict the mobile checkin popularity of regions. Finally, extensive experiments with real-world urban region data demonstrate the effectiveness and efficiency of our method.

This article develops a new 2-D direction estimation algorithm using parallel arrays that consist of two parallel nested subarrays. The key idea behind the presented algorithm is to use the difference coarray properties embedded in the auto- and cross correlation matrices of the parallel nested arrays to form a direction-of-arrival (DOA) matrix with enhanced degree of freedoms for 2-D angle estimation. The presented algorithm does not require 2-D nonlinear searching or parameter paring processing. Simulation results demonstrate the superiority of the new algorithm.

Software-as-a-service (SaaS) has attracted substantial attention as a software delivery and service model in a cloud computing environment. SaaS delivery can help organizations significantly reduce the cost of using software, because the resources for running SaaS applications are shared among tenants (end users or organizations). However, multi-tenancy can considerably complicate SaaS development, deployment, and maintenance as a result of a large number of co-existing tenant-specific constraints or features. Manually configuring and maintaining tenant-specific features will increase the cost, introduce possible errors, and limit scalability and flexibility. The paper addresses the problem of large variations and complex configurations. Specifically, the objective is to develop mechanisms to support automatic multi-tenant software features analysis and matching for the purposes of efficient deployment and operations in the cloud. The emphasis of this paper is on the matching between the tenant-specific requirements and the SaaS features managed by the cloud provider. This paper proposes a novel approach for cloud feature matching using XML filtering techniques to support the process of multi-tenant SaaS deployment and management. Feature modeling has been widely used to capture requirements and constraints. On the other hand, XML filtering techniques are mature and have been adopted in various problem domains. We used Yfilter, a proven and XML filtering technique, to support two multi-tenant applications: (i) Identifying SaaS configurations (in XPath representations) that satisfy tenant-specific requirements and constraints (in XML notation); and (ii) Identifying tenants that have subscribed to a specific set of SaaS features. The applications can effectively facilitate SaaS subsequent management and operations due to various changes, e.g., functionalities, constraints, cost, etc. The experimental results demonstrate that the proposed approach can automatically and correctly identify cloud system configurations that match tenant-specific requirements or identify the group of tenants that have subscribed to a particular set of cloud features. In addition, the execution time of our proposed approach is only a small fraction compared to the existing approach using the formal method, e.g., FaMa, and the configuration space is also much smaller.

In this paper, an improved boundary point interpolation method (BPIM) is developed for solving exterior acoustic radiation problems. With the help of Lagrange basis polynomials, a Lagrange point interpolation method (PIM) is presented to avoid the inversion of the moment matrix in the PIM. Then combining the acoustic double layer potential with the Lagrange PIM, the improved BPIM for exterior acoustic radiation problems is presented. Detailed and efficient computational formulas are established to compute acoustic hypersingular and nearly hypersingular integrals. Numerical results show that the current BPIM can produce efficient numerical solutions to exterior acoustic radiation problems.

Production efficiency can be greatly improved by using batch processing machines which can process several jobs in parallel. When job processing characteristics are different, job family should be further considered to group jobs properly in a batch. The problem of scheduling non-identical jobs from incompatible job families on a batch processing machine is investigated in this paper. Job sizes are non-identical and the objective is to minimize the maximum lateness Lmax. Batch processing time is determined by the job with the longest processing time and batch due date equals to the earliest job due date in the batch. Only jobs from the same job family can be grouped together in the same batch. A mathematical model of the problem under study is formulated and validated by using CPLEX. A lower bound is proposed based on the lower bound and the upper bound of the batch number. Because of the NP-hardness of the problem, heuristics are designed to solve the problem under study. These heuristics are then improved by optimizing completion time or due date of the critical batch. Experimental studies show that heuristics could be effectively improved by CTD (Completion Time Decreasing) rules.

Product promotions and liberal return policies are two effective signals that can increase customers' repurchase behavior when online shopping, and how these signals work at different stages of market growth for various customer groups is an important topic for research and applications. Thus, to help online merchants make more effective decisions regarding the use of such signals, this paper proposes a multi-stage hidden Markov model (MS-HMM) to explore the motivational process behind customer repurchase behavior through the lens of the Signaling Theory. The customer-merchant relationship (CMR) is represented as the latent state in the hidden Markov model and is coupled with stage-heterogeneity in terms of state transition probabilities and state-dependent choice probabilities. Moreover, extensive experiments with real-world data are conducted to validate the effectiveness of the MS-HMM.

A self-adaptive projection method is considered for the numerical simulation of contact problems with Tresca friction. The contact constraints are imposed in the fixed point formulation using projection operators. It leads to a projection method where each iterative step consists of a linear elasticity problem with Robin conditions on the contact boundary. To ensure the efficiency of the method, a self-adaptive rule is given to adjust the parameter automatically. We prove the convergence of the method in function space and demonstrate its application with the boundary element method. The numerical experiments are presented to show the performance of the proposed method.

Image retrieval, which refers to browse, search and retrieve the images of the same scene or object from a large database of digital images, has attracted increasing interests in recent years. This paper proposes a coarse-to-fine method for fast indexing with Deep Convolutional Neural Network(DCNN) and Local Geometrical Constraint Model. We first use a vector quantized DCNN feature descriptors and exploit enhanced Locality-sensitive hashing(LSH) techniques for fast coarse-grained retrieval. Then, we focus on obtaining high-precision preserved matches for fine-grained retrieval. This is formulated as a maximum likelihood estimation of a Bayesian model with latent variables indicating whether matches in the putative set are inliers or outliers. We impose the non-parametric global geometrical constraints on the correspondence using Tikhonov regularizers in a reproducing kernel Hilbert space. To ensure the well-posedness of the problem, we develop a local geometrical constraint that can preserve local structures among neighboring feature points, and it is also robust to a large number of outliers. The problem is solved by using the Expectation Maximization algorithm. Extensive experiments on real near-duplicate images for both feature matching and image retrieval demonstrate that the results of the proposed method outperform current state-of-the-art methods.

The linear threshold model is widely adopted as a classic prototype for studying contagion processes on social networks, where nodes, representing individuals, are assumed to be in one of two states: inactive or active. Each inactive node can be activated via a threshold rule during evolution. Although both contagion mechanisms and network impacts have been well studied, very few studies paid attention on the effect of interacting strengths on the threshold rule. In this paper, a modified linear threshold model on weighted networks is proposed. On one hand, the weight of a link in the network is characterized by a power-law function of the product of endpoint degrees. On the other hand, peer influences on a node incorporate both the number of its active neighbors and associated link weights. The systematic dynamics is explored by the combination of the spin-glass theory and Monte-Carlo algorithm. In analogy to unweighted networks, a global cascade is not triggered in weighted networks when the average degree of nodes is either too small or too large, however, large cascades are realized within an intermediate range, which is referred to as the cascade window. Moreover, two regimes of the power exponent of the weight function are identified in which the system exhibits distinct behaviors: when networks are very sparse, there exist one extreme of the weight exponent making the system susceptible to large cascades; when the networks are relatively dense, on the contrary, there exists the other extreme of the weight exponent causing the system to maintain optimal robustness. All these results demonstrate the importance of both network connectivity and link weights, and offer a sophisticated description of social contagions. Ministry of Education (MOE) This work was partly supported by Natural Science Foundation of China under Grant No. 11331009, Science and Technology Commission of Shanghai Municipality, China under Grant No. 17ZR1445100, and Ministry of Education of Singapore under Grants MOE2014-T2-1-028 and MOE2016-T2-1-119.

A meshless projection iterative method (MPIM) is developed in this paper for solving nonlinear Signorini problems. In the MPIM, a projection iterative method is adopted to linearize the nonlinear Signorini problem to a sequence of linear boundary value problems, and then the moving Kriging interpolation is used to solve these linear problems iteratively. The nonlinear inequality boundary conditions are treated efficiently by the projection iterative method, while the Dirichlet boundary condition is implemented directly with the help of the moving Kriging interpolation. Numerical results demonstrate the efficiency and convergence of the MPIM.

The increasing popularity of social networks has inspired recent research to explore social graphs for data mining. Because social graph data contains sensitive information about users, publishing the graph data directly will cause privacy leakage of users. In this paper, we assume that attackers might re-identify targets with 1-neighborhood graph attacks. To prevent such attacks, we propose a Graph Matching based Privacy-preserving Scheme, named GMPS, to anonymize the social graphs. We utilize Jensen-Shannon Divergence to compute node structure similarity to improve the accuracy of node clustering. And then, utilize the graph modification to achieve k-anonymity and use graph matching to measure the similarity of graphs. The experiment results on HepTh and Facebook show that the proposed approach achieves k-anonymity with low information loss and high data utility.

With the development of China’s economy, some problems such as ecological imbalance, serious pollution of the environment, low utilization of natural resources and so on, are followed. Green economy and green finance are very necessary to alleviate these problems. Based on the annual data of China's green finance and green economy from 2004 to 2017, we build a VAR model to empirically analyze the dynamic relationship between green economy and green finance. The empirical results indicate that increasing investment in green finance positively affects the development of green economy; simultaneously, green economy also has a certain degree of impact on green finance, which has a restraining effect in the early stage, but with the growth of time, green economy is also important to green finance. This study explores the interaction mechanism between green finance and green economy, which is important to promote ecological civilization construction, green economic development and sustainable development.

Live streaming is an effective tool to increase products sales, which has become a new social commerce. However, few studies have explored the psychological mechanisms of individuals’ impulse buying, although live streaming has transformed behaviors of customers. In this study, ground on S-O-R paradigm and motivation theory, we developed a research model to explore how customers develop shopping motivations in live streaming shopping. The results from 318 actual customers in China showed that three situational factors: visibility, parasocial interaction, and social presence of others differently affect form utilitarian and hedonic motivation of consumers. The motivations of both utilitarian and hedonic drive customers’ impulse buying decisions. Specifically, utilitarian motivation plays a less significant role in driving customers’ purchase intention than hedonic motivation. In conclusion, these findings increase one’s comprehension of customers’ psychological mechanisms of impulse buying development in live streaming context and spread the range of motivation theory. The findings also supply perceptions to both sellers and platforms in developing strategies.

Pedestrian re-recognition is a very important research direction in video surveillance. With the emphasis on night video surveillance, pedestrian re-recognition is also being studied from a single mode to a cross-mode direction. Since the images taken by the camera at night are generally divided into two types, thermal imaging and infrared images, corresponding to the RegDB data set and the SYSU-MM01 data set respectively. In order to make the trained model have good performance in both data sets, GAN network is used in this article. The visible light image is generated by CycleGAN to generate the corresponding infrared image, and then the infrared image is generated by PTGAN to generate a thermal imaging style image. Then input the image into the single-stream network for training and learning, and finally optimize the network in an end-to-end manner.

Salient object detection methods focus on detecting precise salient objects and acquiring clear boundaries, but many approaches fail to achieve both of them. To solve this problem, we present A Hybrid Salient Object Detection with Global Context Awareness. The proposed model consists primarily of Multi-level Feature Enhancement (MFE) and Top-layer Processing (TP). MFE module is a module that fuses high-level features, low-level features and corresponding global context information and it is used for foreground enhancement. TP module generates two kinds of features, one is the representative top-layer features, and the other is the global context information that acts on the corresponding MFE module. Comparing our method with other 12 state-of-the-art salient object detection methods with different evaluation metrics, the experimental results show that our method is superior to other methods.

Federated learning enables collaborative machine learning among multiple independent participants while preserving data privacy of each participant through model aggregation during training. However, model aggregation still faces potential risks that are associated with indirect leakage, such as parameters. In this paper, we first propose an algorithm called Meta Operation Neural Network (MONN) to perform basic arithmetic operations on encrypted data and generate operation results in a plaintext way. MONN is actually a general neural network composed of an encoder and a meta operation decoder, where both the encryption and meta decryption are lossless. MONN can be applied to federated learning for secure model aggregation. In this way, data privacy can be well preserved not only from malicious attackers but also untrustworthy servers. Experimental results reveal the following three key properties: 1) The proposed MONN based federated aggregation method (denoted as FedMONN) can reach satisfactory performance comparable with non-federated counterparts; 2) FedMONN is more secure than the classic federated averaging and one-time pad aggregation, even the server is not a trustable third party; 3) FedMONN is much efficient than the federated aggregation based on the Paillier homomorphic encryption technique.

In this paper, we solve the ambulance dispatch problem with a reinforcement learning oriented strategy. The ambulance dispatch problem is defined as deciding which ambulance to pick up which patient. Traditional studies on ambulance dispatch mainly focus on predefined protocols and are verified on simple simulation data, which are not flexible enough when facing the dynamically changing real-world cases. In this paper, we propose an efficient ambulance dispatch method based on the reinforcement learning framework, i.e., Multi-Agent Q-Network with Experience Replay(MAQR). Specifically, we firstly reformulate the ambulance dispatch problem with a multi-agent reinforcement learning framework, and then design the state, action, and reward function correspondingly for the framework. Thirdly, we design a simulator that controls ambulance status, generates patient requests and interacts with ambulances. Finally, we design extensive experiments to demonstrate the superiority of the proposed method.

We study the integration of reinforcement learning and spatial knowledge graph for incremental mobile user profiling, which aims to map mobile users to dynamically-updated profile vectors by incremental learning from a mixed-user event stream. After exploring many profiling methods, we identify a new imitation based criteria to better evaluate and optimize profiling accuracy. Considering the objective of teaching an autonomous agent to imitate a mobile user to plan next-visit based on the user's profile, the user profile is the most accurate when the agent can perfectly mimic the activity patterns of the user. We propose to formulate the problem into a reinforcement learning task, where an agent is a next-visit planner, an action is a POI that a user will visit next, and the state of environment is a fused representation of a user and spatial entities (e.g., POIs, activity types, functional zones). An event that a user takes an action to visit a POI, will change the environment, resulting into a new state of user profiles and spatial entities, which helps the agent to predict next visit more accurately. After analyzing such interactions among events, users, and spatial entities, we identify (1)semantic connectivity among spatial entities, and, thus, introduce a spatial Knowledge Graph (KG) to characterize the semantics of user visits over connected locations, activities, and zones. Besides, we identify (2) mutual influence between users and the spatial KG, and, thus, develop a mutual-updating strategy between users and the spatial KG, mixed with temporal context, to quantify the state representation that evolves over time. Along these lines, we develop a reinforcement learning framework integrated with spatial KG. The proposed framework can achieve incremental learning in multi-user profiling given a mixed-user event stream. Finally, we apply our approach to human mobility activity prediction and present extensive experiments to demonstrate improved performances.

With strict protections and regulations of data privacy and security, conventional machine learning based on centralized datasets is confronted with significant challenges, making artificial intelligence (AI) impractical in many mission-critical and data-sensitive scenarios, such as finance, government, and health. In the meantime, tremendous datasets are scattered in isolated silos in various industries, organizations, different units of an organization, or different branches of an international organization. These valuable data resources are well underused. To advance AI theories and applications, we propose a comprehensive framework (called Knowledge Federation- KF) to address these challenges by enabling AI while preserving data privacy and ownership. Beyond the concepts of federated learning and secure multi-party computation, KF consists of four levels of federation: (1) information level, low-level statistics and computation of data, meeting the requirements of simple queries, searching and simplistic operators; (2) model level, supporting training, learning, and inference; (3) cognition level, enabling abstract feature representation at various levels of abstractions and contexts; (4) knowledge level, fusing knowledge discovery, representation, and reasoning. We further clarify the relationship and differentiation between knowledge federation and other related research areas. We have developed a reference implementation of KF, called iBond Platform, to offer a production-quality KF platform to enable industrial applications in finance, insurance, marketing, and government. The iBond platform will also help establish the KF community and a comprehensive ecosystem and usher in a novel paradigm shift towards secure, privacy-preserving and responsible AI. As far as we know, knowledge federation is the first hierarchical and unified framework for secure multi-party computing (statistics, queries, searching, and low-level operations) and learning (training, representation, discovery, inference, and reasoning).

There are a wide variety kinds of cryptographic devices involved in the space-ground integrated information network systems, which will generate a large amount of data. How to implement efficient storage and processing of cryptographic data in a big data environment is particularly important. According to the overall technical requirements of integration of heaven and earth cross-domain hierarchical password on-demand service, this paper preliminary designed the cryptographic data management platform architecture based on Hadoop. And proposed a solution for monitoring data storage and processing based on Hadoop. Aiming at massive heterogeneous cryptographic data, designed a layered architecture. Improved system data processing speed by using distributed computing processing. In addition, designed the security of the data management platform to solve the bottleneck problem of massive cryptographic data management in the space-ground integrated information network.

Cheap-talk communication between parties with conflicting interests is common in many business and economic settings. Two distinct behavioral economics theories, the trust-embedded model and the level-k model, have emerged to explain how cheap talk works between human decision makers. The trust-embedded model considers that decision makers are motivated by nonpecuniary motives to be trusting and trustworthy. In contrast, the level-k model considers that decision makers are purely self-interested but limited in their ability to think strategically. Although both theories have been successful in explaining cheap-talk behaviors in separate contexts, they point to contrasting drivers for human behaviors. In this paper, we provide the first direct comparison of both theories within the same context. We show that, in a cheap-talk setting that well represents many practical situations, the two models make characteristically distinct and empirically distinguishable predictions. We leverage past experiment data from this setting to determine what aspects of cheap-talk behavior each model captures well and which model (or combination of models) has better explanatory power and predictive performance. We find that the trust-embedded model emerges as the dominant explanation. Our results, thus, highlight the importance of investing in systems and processes to foster trusting and trustworthy relationships in order to facilitate more effective cheap-talk interactions. This paper was accepted by Charles Corbett, operations management.

Variational Autoencoder (VAE) is widely used as a generative model to approximate a model's posterior on latent variables by combining the amortized variational inference and deep neural networks. However, when paired with strong autoregressive decoders, VAE often converges to a degenerated local optimum known as ``posterior collapse''. Previous approaches consider the Kullback–Leibler divergence (KL) individual for each datapoint. We propose to let the KL follow a distribution across the whole dataset, and analyze that it is sufficient to prevent posterior collapse by keeping the expectation of the KL's distribution positive. Then we propose Batch Normalized-VAE (BN-VAE), a simple but effective approach to set a lower bound of the expectation by regularizing the distribution of the approximate posterior's parameters. Without introducing any new model component or modifying the objective, our approach can avoid the posterior collapse effectively and efficiently. We further show that the proposed BN-VAE can be extended to conditional VAE (CVAE). Empirically, our approach surpasses strong autoregressive baselines on language modeling, text classification and dialogue generation, and rivals more complex approaches while keeping almost the same training time as VAE.

Recently web applications have been widely used in enterprises to assist employees in providing effective and efficient business processes. Forecasting upcoming web events in enterprise web applications can be beneficial in many ways, such as efficient caching and recommendation. In this paper, we present a web event forecasting approach, DeepEvent, in enterprise web applications for better anomaly detection. DeepEvent includes three key features: web-specific neural networks to take into account the characteristics of sequential web events, self-supervised learning techniques to overcome the scarcity of labeled data, and sequence embedding techniques to integrate contextual events and capture dependencies among web events. We evaluate DeepEvent on web events collected from six real-world enterprise web applications. Our experimental results demonstrate that DeepEvent is effective in forecasting sequential web events and detecting web based anomalies. DeepEvent provides a context-based system for researchers and practitioners to better forecast web events with situational awareness.

K-means algorithm is a kind of algorithm with higher frequency at present, but with the continuous research, this algorithm gradually exposes some defects in practical application, which is manifested in the choice of random and sensitive noise points, which is difficult. Get accurate and reliable clustering solutions. Aiming at this kind of defect, a K-means algorithm based on the distance density of noise points is proposed. Specifically, it uses the effect of density and distance on the cluster center to perform weighted analysis and prediction of the randomness of its data. Processing, based on the determination of data weights, further introduce the “minimum and maximum principle”, which can automatically select the initial clustering center and determine the number of clustering centers. The research results show that this new algorithm can meet the expected requirements, which can not only ensure the accuracy and reliability of the optimization results from the source, but also help people to conduct a more comprehensive and rigorous analysis of the data.

Learning urban community structures refers to the efforts of quantifying, summarizing, and representing an urban community’s (i) static structures, e.g., Point-Of-Interests (POIs) buildings and corresponding geographic allocations, and (ii) dynamic structures, e.g., human mobility patterns among POIs. By learning the community structures, we can better quantitatively represent urban communities and understand their evolutions in the development of cities. This can help us boost commercial activities, enhance public security, foster social interactions, and, ultimately, yield livable, sustainable, and viable environments. However, due to the complex nature of urban systems, it is traditionally challenging to learn the structures of urban communities. To address this problem, in this article, we propose a collective embedding framework to learn the community structure from multiple periodic spatial-temporal graphs of human mobility. Specifically, we first exploit a probabilistic propagation-based approach to create a set of mobility graphs from periodic human mobility records. In these mobility graphs, the static POIs are regarded as vertexes, the dynamic mobility connectivities between POI pairs are regarded as edges, and the edge weights periodically evolve over time. A collective deep auto-encoder method is then developed to collaboratively learn the embeddings of POIs from multiple spatial-temporal mobility graphs. In addition, we develop a Unsupervised Graph based Weighted Aggregation method to align and aggregate the POI embeddings into the representation of the community structures. We apply the proposed embedding framework to two applications (i.e., spotting vibrant communities and predicting housing price return rates) to evaluate the performance of our proposed method. Extensive experimental results on real-world urban communities and human mobility data demonstrate the effectiveness of the proposed collective embedding framework.

In online social networks, many application problems can be generalized as influence maximization problem, which targets at finding the top- k influential users. Most of the existing influence spread models ignore user’s attitude and interaction and cannot model the dynamic influence process. We propose a novel influence spread model called Fluidspread, using the fluid dynamics theory to reveal the time-evolving influence spread process. In this paper, we model the influence spread process as the fluid update process in three dimensions: the fluid height difference, the fluid temperature and the temperature difference. To the best of our knowledge, this is first attempt of using the fluid dynamics theory in this field. Moreover, we formulate the Maximizing Positive Influenced Users (MPIU) problem and design the Fluidspread greedy algorithm to solve it. Through the experimental results, we demonstrate the effectiveness and efficiency of our Fluidspread model and Fluidspread greedy algorithm.

Feature selection plays a crucial role in machine learning. For evolutionary computation based feature selection, the same particles may be repeatedly generated many times during the population iteration process, and recalculating the fitness values of these particles will cost a large amount of computational resources. Therefore, it is necessary to find these particles fast and skip calculating their fitness to save computation resources. This paper proposes the combination of MD5 encryption algorithm and the competitive swarm optimizer (CSO) algorithm for feature selection. Every particle in the population generated by CSO is encoded with MD5 encryption algorithm. Then the HashMap is used to quickly search for the repetitive particles. Through MD5 coding and HashMap searching, the repetitive particles are found fast, and then the recalculation of the repeated particles are avoid. Experimental results show that our proposed algorithm can significantly reduce the running time of feature selection for low-dimensional data, medium-dimensional data and high-dimensional data. Moreover, our algorithm does not change the framework of the competitive population optimization algorithm, only avoids calculation of the repeatedly generated particles, and therefore has no effect on the accuracy and number of features. Our algorithm can be widely applied to other computational intelligence methods.

Identification of Chinese fir family is a key procedure in forestry cultivation. However, it is time consuming depending on experts' identification. Therefore, in this paper, the method of image based recognition of Chinese fir family is proposed to replace experts' identification. For the cones of different Chinese fir families have different texture features, the textures of the cone is modeled using dominant rotated local binary patterns and these textures is selected by combination of binary competitive swarm optimization algorithm and probabilistic collaborative representation based classifier for filtered out irrelevant, noisy or redundant information. Experimental results demonstrate that the proposed approach is effective in recognizing Chinese fir family especially with feature selection.