Abstract:A Multi-channel graph attention network social recommendation model with disentangling capability was proposed. This model mainly included three modules:the deep clustering module, the aggregation module based on multi-channel graph attention network, and the rating prediction module. Among them, the deep clustering module was used to group users and items. The clustering results can be used to split user-user social graph and user-item interaction graph into multiple subgraph to learn user interest groups and users′ interests in different types of items. The aggregation module learns the attention of different sub-graphs to the prediction results. The rating prediction module input the learned user representation vector and item representation vector into the multilayer perceptron for rating prediction. Extensive experiments on multiple real-world datasets demonstrate that the proposed method is better than other social recommendation algorithms. Specifically, compared with the latest graph neural networks method for social recommendation, the root mean square error is respectively reduced by 2.26% and 2.07% on the Ciao and Epinions datasets, and the mean absolute error is respectively reduced by 2.58% and 3.06%.

Abstract:In order to study the relationship between cluster formation and radius, the Cucker-Smale model with a hierarchical structure of a single leader and the leader speed invariable was considered. The influence of free will on cluster was discussed. The sufficient conditions that the radius has a lower bound (the lower bound was related to the velocity difference, particle number, communication intensity, etc.) were obtained by proving. When the radius was greater than the lower bound, the cluster would be generated. The related conclusion was verified by MATLAB numerical simulation.

Abstract:PCANet (principal component analysis network) is a simple deep learning algorithm with excellent performance in the field of image recognition. Integrating the idea of graph embedding into PCANet, a new image recognition algorithm Smooth-PCANet was proposed. In order to verify the effectiveness of the Smooth-PCANet algorithm, adequate experiments were performed on different data sets such as face, handwritten characters, and images. Compared with several image recognition algorithms based on deep learning, the experiments demonstrated that the Smooth-PCANet achieves higher recognition performance than the PCANet and avoids overfitting more effectively, with a significant advantage in small samples training.

Abstract:In MIML (multi-instance multi-label learning) tasks, labels are often correlated with each other, and DAG (directed acyclic graph) is a common hierarchically structure which often occurs in the prediction of gene ontology biological functions of proteins. Considering the labels with directed acyclic graph structures in MIML, a novel algorithm named MIMLDAG (multi-instance multi-label directed acyclic graph) was proposed. MIMLDAG trained a low-dimensional subspace of shared labels from the feature space of original datasets, minimized the rank loss by a stochastic gradient descent method, and then incorporated the inner DAG hierarchical structure of labels for optimizing the output labels. MIMLDAG was applied to predict the protein functions in multiple datasets, and the results show that MIMLDAG possesses higher efficiency and predictive performance.

Abstract:Oriented to the combinatorial optimization problem with fixed sum of goods and multiple optimal solutions, the problem formulation was given by two examples:the fixed sum real number subset problem and buying wings problem. A integer state and a real number state multi-optimal solution dynamic programming algorithm based on 0-1 decision recursive search was put forward on the foundation of analysis of some classical methods like enumeration. In order to cope with the problem of time complexity tending to the extreme O(mⁿ) when the number of optimal solutions is large for the proposed algorithms, two improved algorithms, the same decision path fusion algorithm and the 0-x decision based algorithm were proposed. The computation time of the improved algorithms is consistent with the proportional relation with O(nb+nm) on the whole in experiments, which indicates that these algorithms have good performance for this type of problem.

Abstract:Urban traffic flow forecasting is of great significance for traffic management and public safety. However, the correlations of traffic raster flow change with time. There are global spatio-temporal correlations in the city, and the contributions of channel-wise features vary on each city region. To tackle these challenges and make more accurate prediction, a novel spatio-temporal neural network model, named 3D-CANet (three-dimensional channel-wise attention network), was designed. A 3D-InnerCA (three-dimensional inner-channel attention) unit was proposed to dynamically capture the global spatio-temporal correlations for different channel-wise features. Meanwhile, an InterCA (inter-channel attention) unit was designed to adaptively recalibrate the contributions of different channel-wise features on each region. The experimental results on three real-world traffic raster flow datasets demonstrate that the predictive performance of the 3D-CANet model was better than the others,which proved the validity of the model proposed.

Abstract:Traditional active learning methods select examples by only considering the predictions of the current model. However, these methods neglect the information of the previous trained models, which reflect the stability of the prediction sequence for each unlabeled example during the active learning stage. Thus, a novel active learning method with instability sampling was proposed, which attempted to estimate the potential utility of each unlabeled examples for improving the model performance based on the difference among predictions of the previous models. The proposed method measured the instability of unlabeled example based on the difference between the posterior probabilities predicted by the previous models, and the example with the largest instability was selected to be queried. Extensive experiments were conducted on multiple datasets with diverse classification models. The experimental results validate the effectiveness of the proposed method.

Abstract:The recognition of individual activities helps in the realisation of functions such as user profiling, personalized recommendations, abnormal behaviour detection, city-wide group behaviour analysis and resource allocation optimisation. A recognition method for the semantics of individual activities based on sparse social media check-in data was proposed. The temporal periodicity and tendency features of activity behaviors were extracted from the check-in data, and a spatial preference quantification algorithm was utilized to extract the preferences of groups and individuals from the spatial relevance between individual and group activities. The natural language embedding model BERT was used to extract the semantics of POIs (point of interest). The temporal features, spatial preference features and text features of POI′s names constituted the joint spatio-temporal features characterizing group and individual preferences, and the joint features were classified by the extreme gradient boosting classifier to obtain the activity semantic recognition results. With the results of comparison experiments and ablation experiments on the Foursquare dataset, it was validated that the model proposed can effectively improve the accuracy of activity semantics recognition.

Abstract:In the construction of the model decision tree, there are many parameters and the parameter combination is complex. The use of grid search and other parameter tuning methods will consume a lot of time, which will affect the improvement of the model performance. A model decision tree with multi-kernel bayesian optimization was proposed. In order to deal with the characteristics of different classified data, three Gaussian processes were used for modeling optimization. The Bayesian optimization technique was used to select the best parameter combination. The experimental results show that the proposed algorithm is better than the traditional model decision tree method in parameter optimization, and can find the global optimal parameter value in the case of a few iterations. To a certain extent, it improves the classification performance of the algorithm and saves a lot of parameter adjustment time.

Abstract:Since the workers have significant differences in the knowledge level and evaluation criteria, the quality of the collected labels varies a lot. It′s of key importance to improve the quality of labels and learning models in crowdsourced label learning. A novel double-confidence inference algorithm was proposed to solve the problem of crowdsourced label inference. The workers′ confidence was obtained via the data distribution characteristics and label information, and then the label was inferred by this confidence so as to improve the quality of the integrated label. The experimental results show that the proposed algorithm outperforms other ground truth inference algorithms only based on label information.

Abstract:Existing deep supervised image hashing approaches rely on substantial labeled image data, which is very difficult to be widely applied in reality. By utilizing tags associated with images as the supervision information, a context-aware deep weakly supervised image hashing method was proposed. The method enhanced the image region representations by adaptively capturing the relevant context information of image region features, and raised the discrimination of the learnt hash codes by introducing a discrimination loss. Extensive experiments on two public datasets show the effectiveness of the method.

Abstract:Stealth aircraft has gradually become a great power, and will continue to play an important role, making stealth technology become the key technology of aircraft design. RCS(radar cross section) measurement of stealth aircraft is a necessary step to design, manufacture and maintain stealth aircraft. The basic process of RCS measurement of stealth aircraft was reviewed from three aspects:RCS test of scaled model, outdoor RCS test of full-scale aircraft and indoor near-field test of full-scale aircraft, the theoretical basis of RCS near-field measurement of stealth aircraft was summarized, and the near-field RCS measurement technology with imaging diagnosis function was analyzed intensively. The development trend and key technologies of stealth aircraft RCS measurement were prospected, which is helpful to have a general understanding of RCS measurement of stealth aircraft and grasp the development direction of RCS measurement.

Abstract:To quantity the comprehensive influence of both random and epistemic uncertainties during the process of flight vehicle stage separation, combined with the characteristics of hybrid model based on random and interval theory, a new reliability assessment method of separation between flight vehicle stages based on the hybrid model of random and interval theory was proposed. According to the requirements of the hypersonic vehicle separation mission, the separation kinetic simulation model was constructed. According to the geometric characteristic of the separation structure between the stages, a new rapid collision detection method was proposed. And the hybrid reliability assessment model of flight vehicle separation mission was constructed. The hybrid reliability assessment model was converted to unconstrained optimization problem of random reliability assessment. Considering of the characteristic of highly nonlinearity of system performance function due to complex external force and moment during the process of flight vehicle stage separation, the unconstrained optimization problem was efficiently solved by efficient global optimization and active learning Kriging method. It is shown that the influence of mixed uncertainty factors on the flight vehicle separation process can be described exactly through this method and the interval of flight vehicle separation mission reliability can be given accurately, which can further support the detailed design of flight vehicle separation.

Abstract:Cohesive element is an important means to investigate the "dewetting" damage of propellant. The high filling ratio geometric model of propellant was constructed by the molecular dynamics method, and the mesoscopic finite element analysis model of propellant was constructed by combining the periodic geometry and periodic boundary treatment methods. The "dewetting" behavior of the interface between the particle and the matrix was simulated by using the cohesion element and PPR cohesive zone model. The mechanical response of propellant mesoscopic structure was analyzed under the uniaxial tensile and pure shear tests, and the damage mechanism of "dewetting" of propellant was studied. According to different volume fraction ratio, strain rate and cohesive strength, the influence law of "dewetting" damage was analyzed. The research methods and conclusions can provide a useful reference for the formulation of a new generation of high-performance propellant.

Abstract:Developing the cluster parachute deceleration landing system is a key to realize the requirements of efficient deceleration and lossless landing of heavy-duty spacecraft. Based on aerodynamic theory, simulation analysis was performed on aerodynamic performance of single parachute in the parachute system. Taking simulation results as a reference, the configuration design of large ring sail umbrella was optimized..Based on the correctness and convergence of the simulation method, the cluster parachute system with multiple single parachute configurations was regarded as a research object, the aerodynamic resistance performance and attitude stability of the parachute system when the canopy is full were studied Through the flow field analysis and data comparison of the simulation results, it is found that the parachute with window and gap can not only ensure the good resistance characteristics of the cluster parachute system, but also maintain the working stability of the parachute system. This design can avoid the collision between single parachute due to force.

Abstract:Aiming at the charging process of the spacecraft in the solar wind plasma, based on the spacecraft surface charging simulation software developed by the European Space Agency, a numerical calculation model of the spacecraft was established. The interaction between the solar wind plasma and the spacecraft was simulated, and different surface materials were simulated. The calculation and analysis of the surface potential of the spacecraft and the structure of the wake were realized, and the spatial potential structure and plasma density distribution around the spacecraft in the near-day environment were given. The research results show that the choice of surface material will affect the depth of the space barrier and the structure of the wake, and the surface material whose surface is charged to a lower potential has a smaller wake area; the potential distribution of the spacecraft towards the sun is mainly dominated by photoelectrons and secondary electrons, while the back of the potential distribution of the positive side is mainly affected by the wake effect. The above results have certain theoretical reference value for related work such as detection and evaluation of solar wind environment spacecraft and engineering research.

Abstract:In order to obtain the effect law and genesis mechanism of fuel injection advance angle on the performance of diesel engine under the adjustable fuel injection law conditions, on the basis of the realization of adjustable fuel injection law injection, the working process calculation model of diesel engine was established by Fire software, and the main parameters of model were calibrated through experiments, then the performance of diesel engine with the fuel injection advance angle under saddle-shaped fuel injection law was analyzed with the model. The results show that with the delay of pressurization time, the fuel injection law changes from rectangle to slope and then to saddle-shaped, which realizes the flexible and adjustable fuel injection law. Under the condition of saddle-shaped fuel injection law, there is an optimal fuel injection advance angle to optimize the power and economy of diesel engine. At the same time, with the increase of fuel injection advance angle, the in-cylinder pressure, in-cylinder temperature and heat release rate of diesel engine increase gradually, and the time of reaching their peak values all move forward, while the NO_x emissions and soot emissions show an increasing and decreasing trend respectively.

Abstract:In order to study the influence of TA2-B10 alloy tube on the corrosion characteristics of B10 tube under different galvanic corrosion protection methods, the direct connection, electrical insulation connection and electrical insulation + coating of TA2-B10 tube with a diameter of DN80 were set up in Qingdao Maidao seawater test site. After connect three groups of control pipelines with different galvanic corrosion protection methods, flowing seawater and immersion alternate corrosion tests of 1 m/s, 3 m/s, and 4 m/s were conducted in sequence.After test, three sets of B10 pipelines were cut. Through the pipeline inner surface potential distribution test, the corrosion type of the B10 pipeline under different galvanic corrosion protection methods were analyzed; the potential polarization curve, electrochemical impedance spectroscopy and microscopic characterization were used to analyze the corrosion characteristics of the B10 at a distance of 250 mm from the flange joint under different galvanic corrosion protection methods. Results show that the directly connected TA2-B10 tube is in a galvanic corrosion state, the potential in the B10 terminal is positively shifted and the corrosion is accelerated, and the electrical insulation connection and the insulation+coating connection TA2-B10 tube are in a self-corrosive state. The electrical insulation+coating connection has better insulation effect; the B10 sample under electrical insulation+coating connection protection has the smallest corrosion current density and the most negative self-corrosion potential; the impedance spectra of the three groups of B10 samples all show the characteristics of single capacitive impedance arc, the B10 sample of electrical insulation+coating under the protection of layer connection has greater transmission resistance and film layer resistance; electrical insulation+coating connection can effectively slow down the three corrosion tendencies of pitting, pointed and intergranular corrosion. The above results comprehensively show that the insulation+coating protection method has a better galvanic corrosion protection effect.

Abstract:A mathematical calculation model of armature inductance gradient under the conditions of uniform and non-uniform magnetic field distribution between guide rails was deduced. The velocity frequency was introduced to simulate the velocity skin effect of armature emission process, and the time-harmonic simulation analysis was carried out on the 2D and the 3D electromagnetic field model. The inductance coefficient and the inductance gradient obtained from the simulation were used to calculate the dynamic inductance and the dynamic armature force in the simulation system. Electrical simulation and experiment results show that both the current and the velocity error are less than 2%, which proves the correctness and accuracy of dynamic inductance gradient analysis and parameter extraction method.

Abstract:Aiming at the problem of difficult and limited sample acquisition in SAR(synthetic aperture radar) image target detection, a learning model of joint GAN (generative adversarial network) and detection network was proposed. The original training set was used to pretrain the specially designed faster regional convolutional neural network. The deep convolutional GAN based on attention mechanism was employed to generate extensive synthetic samples, which were input into the detection network for prediction. The corresponding annotation information of the new samples was determined by the prediction information and probability equivalent class label allocation strategy, and the annotated new samples were used to expand the original dataset with a certain proportion. The detection network was retrained with the expanded dataset. Simulation results show that the proposed framework can improve the detection efficiency and performance of the network effectively.

Abstract:In order to quickly evaluate the underwater corrosion electric field in the damaged state of the submarine coating, the corrosion current equivalent circuit of submarine coating local damage was established based on the principle of electrochemical corrosion and the characteristics of submarine structure, and the corrosion current intensity of submarine coating local damage was calculated. The corrosion electric field of submarine was modeled based on the point current source. The damaged part of submarine coating and exposed propeller were equivalent to point current source. The calculation formula of electric field of point current source in layered medium was used to estimate the corrosion stable electric field in the damaged state of the submarine coating. Compared with the simulation results of the commercial finite element software COMSOL for the corrosion electric field of a certain submarine, the results show that the submarine surface corrosion current and the electric field distribution curves of different paths obtained by this estimation method are basically consistent with the simulation results of COMSOL, the relative error of current calculation value does not exceed 6.5%,and the relative error of peak to peak value of each electric field component does not exceed 18%.

Abstract:Packet classification is the fundamental function of network, and researchers have proposed many packet classification solutions in the past two decades. Among them, the decision tree algorithm for packet classification has received extensive attention and in-depth research due to its high throughput, suitable for multiple fields and pipelining. The recent research on the decision tree algorithm for packet classification was introduced, the geometric meaning, common techniques and test benchmarks of the decision tree algorithm were described, and the decision tree algorithm from the two dimensions of node cutting technology and rule set grouping technology were systematically analyzed. The typical algorithms of the two types of common technologies for building decision tree were introduced respectively, the design ideas and characteristics of various algorithms were compared, and their applicable scenarios were given. The conclusion and discuss the future work of decision tree algorithms were stated out.

Abstract:Based on the GBS(graph-based segmentation) algorithm and the hierarchical clustering algorithm, a method to solve the under-segmentation of GBS algorithm was constructed. Meanwhile, the way of multi-threaded parallel processing of data was used to effectively improve the processing speed of the traditional hierarchical clustering algorithm. In the RGB color space, the GBS algorithm was used to obtain the initial segmentation result of each pixel in the image. The pixel value in each type of region was extracted and the hierarchical clustering was carried out to obtain the category label of pixel value in each type of region. According to the category label obtained by hierarchical clustering and the preset category range, the initial segmentation result of each pixel was modified. A new segmentation graph was generated according to the region merging criterion. Experimental results show that compare with the K-means-SLIC algorithm and the GBS algorithm, this method solves the phenomenon of under-segmentation, and produces a semantic segmentation graph with high segmentation accuracy.

Abstract:Aiming at the problem that the prominent features of hesitancy and fuzziness of decision information and it′s difficult to use the traditional method to determine the varieties of spare parts for new equipment of small scale, a method of determining the varieties of spare parts based on hesitating fuzzy rough sets was proposed. The risk preference coefficient was used to extend the incomplete hesitancy information, which laid a foundation for establishing the hesitancy information system for different risk preference. Considering the influence of the score function and the numerical continuation boundary, an improved inclusion degree formula was given and proved on the basis of the definition of inclusion degree. The reduction condition and rule acquisition method of spare parts varieties decision attribute based on the improved inclusion calculation were given, which realized the depth mining and effective utilization of hesitancy and decision information. The results show that this method can deal with hesitancy and decision information effectively, and obtain a simplified and practical decision rule set of spare parts varieties, and the feasibility of the method was verified.